AIPUB归智期刊联盟
Mobile website
- Latest News
- Recommended
- Deep Reinforcement Learning Cluster Scheduling Optimization Based on Improved Meta-learning
- Spectral Knowledge Distillation with Integrated Feature Enhancement
- Advances in Active Learning Algorithms for Timed Automata
- Application Progress of Deep Learning for Breast Cancer Focus Image Segmentation
- Review of Multi-agent Path Finding
- Construction and Application of Carbon Knowledge Base Driven by Large Language Model More+
- Current Issue
- Online First
- Archive
- Click Rank
- Most Downloaded
- 综述文章
- Monographic Series
Article Search
Search by issue
Select AllDeselectExport
Display Method:
2026,35(1):1-18, DOI: 10.15888/j.cnki.csa.010074, CSTR: 32024.14.csa.010074
Abstract:
Multi-agent path finding (MAPF) aims to plan conflict-free paths for multiple agents to optimize collaborative task performance. This study reviews the current state of MAPF research, including algorithm classification, application scenarios, and future trends, while discussing the challenges in large-scale dynamic environments. First, the study provides a detailed introduction to the definition of MAPF. Then, it categorizes and summarizes path planning algorithms based on search, bio-inspired methods, sampling, and reinforcement learning. Finally, the study analyzes the advantages and disadvantages of each algorithm and their applicable scenarios. This review aims to help researchers understand the current developments and future directions of MAPF technology, and to promote further progress in this field.
Multi-agent path finding (MAPF) aims to plan conflict-free paths for multiple agents to optimize collaborative task performance. This study reviews the current state of MAPF research, including algorithm classification, application scenarios, and future trends, while discussing the challenges in large-scale dynamic environments. First, the study provides a detailed introduction to the definition of MAPF. Then, it categorizes and summarizes path planning algorithms based on search, bio-inspired methods, sampling, and reinforcement learning. Finally, the study analyzes the advantages and disadvantages of each algorithm and their applicable scenarios. This review aims to help researchers understand the current developments and future directions of MAPF technology, and to promote further progress in this field.
2026,35(1):19-38, DOI: 10.15888/j.cnki.csa.010029, CSTR: 32024.14.csa.010029
Abstract:
As one of the most common malignant tumors with high mortality rates among women worldwide, breast cancer relies highly on medical imaging for its diagnosis and treatment. Focus segmentation plays a crucial role in the precise identification of pathological regions, diagnostic assistance, and treatment planning. Recent advances in deep learning have made significant progress in automatic segmentation for breast cancer focus, laying a foundation for further deep learning-based research in this field. Recent research achievements are systematically reviewed, with a focus on the application of deep learning technology in focus segmentation under different medical imaging modes, aiming to provide a reference for the advancement of research in breast cancer focus segmentation. The relevant datasets and common evaluation metrics for image segmentation are briefly introduced. The image segmentation method for breast cancer based on deep learning is then systematically reviewed. The application of algorithms applied in different imaging modes is generalized. At last, current challenges for this technology are summarized. Future directions are also discussed based on limitations in the existing research.
As one of the most common malignant tumors with high mortality rates among women worldwide, breast cancer relies highly on medical imaging for its diagnosis and treatment. Focus segmentation plays a crucial role in the precise identification of pathological regions, diagnostic assistance, and treatment planning. Recent advances in deep learning have made significant progress in automatic segmentation for breast cancer focus, laying a foundation for further deep learning-based research in this field. Recent research achievements are systematically reviewed, with a focus on the application of deep learning technology in focus segmentation under different medical imaging modes, aiming to provide a reference for the advancement of research in breast cancer focus segmentation. The relevant datasets and common evaluation metrics for image segmentation are briefly introduced. The image segmentation method for breast cancer based on deep learning is then systematically reviewed. The application of algorithms applied in different imaging modes is generalized. At last, current challenges for this technology are summarized. Future directions are also discussed based on limitations in the existing research.
2026,35(1):39-51, DOI: 10.15888/j.cnki.csa.010055, CSTR: 32024.14.csa.010055
Abstract:
As an important formal tool for describing the time-constrained behavior of real-time systems, timed automata are widely employed in fields such as embedded systems and communication protocols. The traditional way of manually building real-time system models is time-consuming and prone to errors, and automatic inference models have become a research hotspot. This study focuses on the active learning algorithms of time automata, sorts them out according to the data storage structure and equivalent query method, summarizes both the latest research status of active learning algorithms in the current field of time automata, and their core ideas and technical frameworks, with the challenges faced by the current research analyzed at the same time. By comparing the advantages and limitations of various methods, this study hopes to provide researchers with a clear reference framework and propose possible future research ideas, aiming to promote the development of the theory and practice of TA automated modeling.
As an important formal tool for describing the time-constrained behavior of real-time systems, timed automata are widely employed in fields such as embedded systems and communication protocols. The traditional way of manually building real-time system models is time-consuming and prone to errors, and automatic inference models have become a research hotspot. This study focuses on the active learning algorithms of time automata, sorts them out according to the data storage structure and equivalent query method, summarizes both the latest research status of active learning algorithms in the current field of time automata, and their core ideas and technical frameworks, with the challenges faced by the current research analyzed at the same time. By comparing the advantages and limitations of various methods, this study hopes to provide researchers with a clear reference framework and propose possible future research ideas, aiming to promote the development of the theory and practice of TA automated modeling.
2026,35(1):52-63, DOI: 10.15888/j.cnki.csa.010043, CSTR: 32024.14.csa.010043
Abstract:
Time series forecasting finds widespread applications in such fields as weather forecasting, power load forecasting, and financial management. In recent years, deep learning has made remarkable progress in these tasks. However, existing models still have limitations in struggling with non-stationarity and heterogeneous pattern modeling, which is mainly represented by homogenized modeling of trends and seasonal components, and modal aliasing during decomposition. To this end, this study proposes a time-frequency cooperative decomposition network, termed as DTFNet, which designs a heterogeneous architecture with parallel time and frequency domains. In the time domain, an MLP network with strong noise resistance is employed to model the long-term evolution characteristics of trends, while in the frequency domain, fast Fourier transform is adopted to extract periodic seasonal components, with multi-scale convolution operations employed to capture spatial correlations between time-frequency characteristics. Meanwhile, this study introduces a decomposition method based on discrete wavelet transformation (DWT) to replace conventional moving average decomposition, effectively mitigating boundary effects and modal aliasing. Experiments on six public datasets demonstrate that DTFNet outperforms the current mainstream models in both accuracy and robustness. Ablation experiments show the notable effectiveness of the proposed DWT-based decomposition module and dual-domain time-frequency modeling architecture. Featuring sound generalization ability, DTFNet is applicable to multiple time series forecasting tasks, offering powerful support for real-world applications such as power load forecasting and weather forecasting.
Time series forecasting finds widespread applications in such fields as weather forecasting, power load forecasting, and financial management. In recent years, deep learning has made remarkable progress in these tasks. However, existing models still have limitations in struggling with non-stationarity and heterogeneous pattern modeling, which is mainly represented by homogenized modeling of trends and seasonal components, and modal aliasing during decomposition. To this end, this study proposes a time-frequency cooperative decomposition network, termed as DTFNet, which designs a heterogeneous architecture with parallel time and frequency domains. In the time domain, an MLP network with strong noise resistance is employed to model the long-term evolution characteristics of trends, while in the frequency domain, fast Fourier transform is adopted to extract periodic seasonal components, with multi-scale convolution operations employed to capture spatial correlations between time-frequency characteristics. Meanwhile, this study introduces a decomposition method based on discrete wavelet transformation (DWT) to replace conventional moving average decomposition, effectively mitigating boundary effects and modal aliasing. Experiments on six public datasets demonstrate that DTFNet outperforms the current mainstream models in both accuracy and robustness. Ablation experiments show the notable effectiveness of the proposed DWT-based decomposition module and dual-domain time-frequency modeling architecture. Featuring sound generalization ability, DTFNet is applicable to multiple time series forecasting tasks, offering powerful support for real-world applications such as power load forecasting and weather forecasting.
2026,35(1):64-75, DOI: 10.15888/j.cnki.csa.010080, CSTR: 32024.14.csa.010080
Abstract:
Road defect detection, as an important method for measuring pavement damage and maintaining road maintenance, faces challenges, including extreme length-to-width ratios, varying defect sizes, and uneven distributions of easy versus difficult defects. Current convolution-based methods have achieved larger receptive fields to enhance perception, but at the expense of high-frequency components that contain small defects, making them unsuitable for road defect detection tasks. To address this, a road defect detection algorithm, FS-YOLO, based on frequency enhancement and synergy of geometric shape and category, is proposed. First, to balance the receptive field and high-frequency information, a frequency-adaptive dilation strategy is introduced, dynamically adjusting the spatial expansion rate according to local frequency components, and assigning appropriate convolutional kernels to defects of different sizes. Second, given that different types of defects have distinct geometric shapes and positions, an attention-based three-dimensional explicit synergy dynamic detection head is introduced to achieve explicit synergy between spatial geometric information and category information, enabling the model to leverage the inherent potential of defect categories and spatial locations. Finally, the Slide loss function is introduced to address the imbalance in the distribution of difficult and easy defects in real-world roads, particularly enhancing the model’s ability to handle difficult-to-distinguish samples. Experimental results show that FS-YOLO significantly outperforms the baseline model in terms of precision and recall on both the self-built dataset and the public road defect detection datasets RDD 2022 and UAV-PDD. It has also been effectively validated in practical applications on expressways and national and provincial roads, significantly improving the accuracy and efficiency of defect detection.
Road defect detection, as an important method for measuring pavement damage and maintaining road maintenance, faces challenges, including extreme length-to-width ratios, varying defect sizes, and uneven distributions of easy versus difficult defects. Current convolution-based methods have achieved larger receptive fields to enhance perception, but at the expense of high-frequency components that contain small defects, making them unsuitable for road defect detection tasks. To address this, a road defect detection algorithm, FS-YOLO, based on frequency enhancement and synergy of geometric shape and category, is proposed. First, to balance the receptive field and high-frequency information, a frequency-adaptive dilation strategy is introduced, dynamically adjusting the spatial expansion rate according to local frequency components, and assigning appropriate convolutional kernels to defects of different sizes. Second, given that different types of defects have distinct geometric shapes and positions, an attention-based three-dimensional explicit synergy dynamic detection head is introduced to achieve explicit synergy between spatial geometric information and category information, enabling the model to leverage the inherent potential of defect categories and spatial locations. Finally, the Slide loss function is introduced to address the imbalance in the distribution of difficult and easy defects in real-world roads, particularly enhancing the model’s ability to handle difficult-to-distinguish samples. Experimental results show that FS-YOLO significantly outperforms the baseline model in terms of precision and recall on both the self-built dataset and the public road defect detection datasets RDD 2022 and UAV-PDD. It has also been effectively validated in practical applications on expressways and national and provincial roads, significantly improving the accuracy and efficiency of defect detection.
2026,35(1):76-87, DOI: 10.15888/j.cnki.csa.010040, CSTR: 32024.14.csa.010040
Abstract:
Existing generative adversarial network (GAN) compression methods focus more on optimizing network architecture and the spatial domain, while neglecting the impact of spectral-domain optimization on distillation effectiveness and model performance. This limitation results in discrepancies between lightweight models and teacher models in generating high-frequency image details. In addition, conventional feature extraction methods in image translation often cause detail loss. To address these issues, this study proposes a spectral knowledge distillation scheme with integrated feature enhancement (FESD-CycleGAN). In FESD-CycleGAN, by shifting certain feature channels in the feature map, the receptive field is expanded and feature diversity is enhanced, thus improving both the details and the overall quality of generated images. Moreover, since spectral-domain knowledge distillation enables the generator to capture high-frequency details of images, knowledge distillation in both the spatial and spectral domains is integrated on top of feature enhancement in the feature map. This approach enhances the model’s ability to preserve fine details in generated images. Experimental results show that on the horse2zebra, summer2winter, and edges2shoes datasets, FESD-CycleGAN reduces the FID by 2.19, 0.68, and 0.76 compared to the baseline DCD model, achieving scores of 54.98, 73.41, and 27.45, respectively. The generative performance of lightweight models is effectively improved by FESD-CycleGAN.
Existing generative adversarial network (GAN) compression methods focus more on optimizing network architecture and the spatial domain, while neglecting the impact of spectral-domain optimization on distillation effectiveness and model performance. This limitation results in discrepancies between lightweight models and teacher models in generating high-frequency image details. In addition, conventional feature extraction methods in image translation often cause detail loss. To address these issues, this study proposes a spectral knowledge distillation scheme with integrated feature enhancement (FESD-CycleGAN). In FESD-CycleGAN, by shifting certain feature channels in the feature map, the receptive field is expanded and feature diversity is enhanced, thus improving both the details and the overall quality of generated images. Moreover, since spectral-domain knowledge distillation enables the generator to capture high-frequency details of images, knowledge distillation in both the spatial and spectral domains is integrated on top of feature enhancement in the feature map. This approach enhances the model’s ability to preserve fine details in generated images. Experimental results show that on the horse2zebra, summer2winter, and edges2shoes datasets, FESD-CycleGAN reduces the FID by 2.19, 0.68, and 0.76 compared to the baseline DCD model, achieving scores of 54.98, 73.41, and 27.45, respectively. The generative performance of lightweight models is effectively improved by FESD-CycleGAN.
Article Search
Search by issue
Select AllDeselectExport
Display Method:
Available online: January 19, 2026
,DOI: 10.15888/j.cnki.csa.010090
Abstract:
In 2018, the researchers from the Massachusetts Institute of Technology, inspired by the neural network of Caenorhabditis elegans, proposed the liquid neural network (LNN). This type of neural network more closely resembles the thinking patterns of the human brain and can process sequential tasks more efficiently. This study introduces and analyzes research related to LNN. It primarily summarizes the principal models of LNN, highlighting their distinctions from and connections to simple recurrent neural network (Simple RNN), long short-term memory (LSTM) network, and time-constant recurrent neural network (TC-RNN), as well as the advantages that LNN possesses over TC-RNN. Furthermore, it details the applications of LNN in autonomous driving, drone navigation, and stock prediction, analyzing the specific LNN models employed in these areas. Finally, the study summarizes the challenges faced and discusses the prospects for future development.
In 2018, the researchers from the Massachusetts Institute of Technology, inspired by the neural network of Caenorhabditis elegans, proposed the liquid neural network (LNN). This type of neural network more closely resembles the thinking patterns of the human brain and can process sequential tasks more efficiently. This study introduces and analyzes research related to LNN. It primarily summarizes the principal models of LNN, highlighting their distinctions from and connections to simple recurrent neural network (Simple RNN), long short-term memory (LSTM) network, and time-constant recurrent neural network (TC-RNN), as well as the advantages that LNN possesses over TC-RNN. Furthermore, it details the applications of LNN in autonomous driving, drone navigation, and stock prediction, analyzing the specific LNN models employed in these areas. Finally, the study summarizes the challenges faced and discusses the prospects for future development.
Available online: January 19, 2026
,DOI: 10.15888/j.cnki.csa.010098
Abstract:
Aiming to address the problems of traditional image generation models, including insufficient latent space representation capability in complex scenes and low fidelity in high-resolution image generation, this study proposes a two-stage training framework based on an improved vector quantized variational autoencoder (IVQ-VAE) and a feature-fused Transformer diffusion (FFTD) model. By introducing the attention mechanism, residual blocks, and multi-component loss function, IVQ-VAE significantly enhances the semantic representation capability of the latent space and fidelity of generated images, overcoming the limitations of traditional encoders in capturing complex image features. Built on a Transformer architecture, FFTD further improves the modeling capacity of complex image structures by integrating multi-resolution sampling and adaptive feature fusion. The two-stage training strategy first pre-trains IVQ-VAE to generate high-quality latent representations, then freezes its parameters, and trains FFTD by employing a denoising diffusion implicit model (DDIM) to optimize the noise prediction and image generation process. This framework achieves significant improvements in detail fidelity and visual quality of the generated images on datasets such as CelebA-HQ and AFHQ, validating its effectiveness in high-resolution image generation.
Aiming to address the problems of traditional image generation models, including insufficient latent space representation capability in complex scenes and low fidelity in high-resolution image generation, this study proposes a two-stage training framework based on an improved vector quantized variational autoencoder (IVQ-VAE) and a feature-fused Transformer diffusion (FFTD) model. By introducing the attention mechanism, residual blocks, and multi-component loss function, IVQ-VAE significantly enhances the semantic representation capability of the latent space and fidelity of generated images, overcoming the limitations of traditional encoders in capturing complex image features. Built on a Transformer architecture, FFTD further improves the modeling capacity of complex image structures by integrating multi-resolution sampling and adaptive feature fusion. The two-stage training strategy first pre-trains IVQ-VAE to generate high-quality latent representations, then freezes its parameters, and trains FFTD by employing a denoising diffusion implicit model (DDIM) to optimize the noise prediction and image generation process. This framework achieves significant improvements in detail fidelity and visual quality of the generated images on datasets such as CelebA-HQ and AFHQ, validating its effectiveness in high-resolution image generation.
Available online: January 19, 2026
,DOI: 10.15888/j.cnki.csa.010102
Abstract:
In the medical field, retrieval-augmented generation (RAG) has been proposed to mitigate hallucinations in large language model (LLM) and enhance the interpretability and controllability. However, existing techniques are faced with poor recall of low-frequency entities and difficulties in processing ambiguous, verbose, or polysemous queries. To this end, this study proposes an iterative hybrid retrieval-augmented generation (IHRAG) approach for LLM to improve the intention parsing ability of complex queries and enhance the model’s performance in knowledge mining capabilities for making LLMs generate more accurate responses. IHRAG employs a dynamic routing mechanism to synergistically leverage the semantic generalization capability of vector retrieval andthe structured reasoning capacity of knowledge graphs. By combining a medical ontology-driven query decomposition algorithm, complex clinical questionsare broken down into retrievable atomic sub-questions. Furthermore, a knowledge gap-aware neuro-symbolic expansion model and a “retrieve-verify-iterate” closed-loop optimization mechanism are introduced to establish a progressive discovery process that advances from surface-level information extraction to deep knowledge mining. Experiments demonstrate that IHRAG significantly enhances the performance of base models of various scales such as Qwen and DeepSeek, achieving an improvement in the accuracy of up to 11.12 percentage points and a 17% increase in the high-quality response rate.
In the medical field, retrieval-augmented generation (RAG) has been proposed to mitigate hallucinations in large language model (LLM) and enhance the interpretability and controllability. However, existing techniques are faced with poor recall of low-frequency entities and difficulties in processing ambiguous, verbose, or polysemous queries. To this end, this study proposes an iterative hybrid retrieval-augmented generation (IHRAG) approach for LLM to improve the intention parsing ability of complex queries and enhance the model’s performance in knowledge mining capabilities for making LLMs generate more accurate responses. IHRAG employs a dynamic routing mechanism to synergistically leverage the semantic generalization capability of vector retrieval andthe structured reasoning capacity of knowledge graphs. By combining a medical ontology-driven query decomposition algorithm, complex clinical questionsare broken down into retrievable atomic sub-questions. Furthermore, a knowledge gap-aware neuro-symbolic expansion model and a “retrieve-verify-iterate” closed-loop optimization mechanism are introduced to establish a progressive discovery process that advances from surface-level information extraction to deep knowledge mining. Experiments demonstrate that IHRAG significantly enhances the performance of base models of various scales such as Qwen and DeepSeek, achieving an improvement in the accuracy of up to 11.12 percentage points and a 17% increase in the high-quality response rate.
Available online: January 19, 2026
,DOI: 10.15888/j.cnki.csa.010103
Abstract:
Molecular property prediction and compound-protein interaction (CPI) prediction are key steps in drug discovery. However, traditional graph convolutional network (GCN) are limited by local receptive fields and cannot fully capture the complexity of chemical structures, dynamic changes of molecular conformation, and long-range electronic interactions, which causes bottlenecks to the prediction performance. To this end, this study proposes a deep learning model, T-SeGAT, designed to improve the accuracy and generalization ability of molecular property and CPI prediction. T-SeGAT integrates the ESM-2 protein language model, ChemBERTa molecular language model, and a graph neural network based on graph attention network (GAT) and Set2Set, thereby enabling multi-level feature extraction and fusion from sequence to structure. Meanwhile, to handle the imbalance of experimental data, the model introduces weighted random sampling, balanced/focal/adaptive loss functions, and a dynamic threshold search mechanism at the levels of data loading, loss calculation, and prediction decision-making. Furthermore, it combines an AUC difference-based overfitting suppression method, early stopping strategy, and learning rate scheduling to enhance training stability and generalization ability. Experiments are conducted on the BACE, P53, and hERG datasets for molecular property prediction, and on the Human and C. elegans datasets for CPI prediction, with stratified five-fold cross-validation adopted for performance evaluation. The results show that T-SeGAT consistently outperforms existing baseline models on all datasets. Among them, on the BACE and hERG datasets, the AUC and precision improved by 0.022, 0.010 and 0.004, 0.022 respectively compared with the second-best model, while on the Human dataset, precision increases by 0.013. In conclusion, T-SeGAT demonstrates clear advantages in accuracy, stability, and practicality, providing powerful support for molecular property and CPI prediction in drug discovery.
Molecular property prediction and compound-protein interaction (CPI) prediction are key steps in drug discovery. However, traditional graph convolutional network (GCN) are limited by local receptive fields and cannot fully capture the complexity of chemical structures, dynamic changes of molecular conformation, and long-range electronic interactions, which causes bottlenecks to the prediction performance. To this end, this study proposes a deep learning model, T-SeGAT, designed to improve the accuracy and generalization ability of molecular property and CPI prediction. T-SeGAT integrates the ESM-2 protein language model, ChemBERTa molecular language model, and a graph neural network based on graph attention network (GAT) and Set2Set, thereby enabling multi-level feature extraction and fusion from sequence to structure. Meanwhile, to handle the imbalance of experimental data, the model introduces weighted random sampling, balanced/focal/adaptive loss functions, and a dynamic threshold search mechanism at the levels of data loading, loss calculation, and prediction decision-making. Furthermore, it combines an AUC difference-based overfitting suppression method, early stopping strategy, and learning rate scheduling to enhance training stability and generalization ability. Experiments are conducted on the BACE, P53, and hERG datasets for molecular property prediction, and on the Human and C. elegans datasets for CPI prediction, with stratified five-fold cross-validation adopted for performance evaluation. The results show that T-SeGAT consistently outperforms existing baseline models on all datasets. Among them, on the BACE and hERG datasets, the AUC and precision improved by 0.022, 0.010 and 0.004, 0.022 respectively compared with the second-best model, while on the Human dataset, precision increases by 0.013. In conclusion, T-SeGAT demonstrates clear advantages in accuracy, stability, and practicality, providing powerful support for molecular property and CPI prediction in drug discovery.
Available online: January 19, 2026
,DOI: 10.15888/j.cnki.csa.010107
Abstract:
Traditional multimodal sentiment analysis methods often suffer from information redundancy during feature concatenation and fusion, making it difficult to capture fine-grained and complex emotional features, while also exhibiting limited robustness in modality-missing and cross-domain transfer scenarios. Meanwhile, most existing mixture of experts (MoE) methods adopt a single-layered structure with ambiguous expert specialization, leading to functional overlap and suboptimal generalization. To address these issues, this study proposes a hierarchical gated expert mixture (H-GEM) model. A three-layer hierarchical expert architecture is constructed: a modality expert layer extracts modal features, a fusion and abstraction expert layer adaptively selects fusion strategies, and a sentiment polarity expert layer performs fine-grained modeling. In addition, information-theoretic and discriminative constraints are incorporated to enhance the semantic discriminability and sparsity of expert selection. By leveraging hierarchical gating for progressive decision-making, H-GEM ensures differentiated expert specialization and cross-task modeling. Experiments on CMU-MOSI and CMU-MOSEI datasets demonstrate that H-GEM outperforms baseline models across a series of metrics. Compared with single-layer MoE architectures, the significantly reduced routing entropy indicates effective mitigation of expert redundancy. Moreover, the proposed model demonstrates higher robustness in low-resource and modality-missing scenarios, highlighting its strong practical applicability.
Traditional multimodal sentiment analysis methods often suffer from information redundancy during feature concatenation and fusion, making it difficult to capture fine-grained and complex emotional features, while also exhibiting limited robustness in modality-missing and cross-domain transfer scenarios. Meanwhile, most existing mixture of experts (MoE) methods adopt a single-layered structure with ambiguous expert specialization, leading to functional overlap and suboptimal generalization. To address these issues, this study proposes a hierarchical gated expert mixture (H-GEM) model. A three-layer hierarchical expert architecture is constructed: a modality expert layer extracts modal features, a fusion and abstraction expert layer adaptively selects fusion strategies, and a sentiment polarity expert layer performs fine-grained modeling. In addition, information-theoretic and discriminative constraints are incorporated to enhance the semantic discriminability and sparsity of expert selection. By leveraging hierarchical gating for progressive decision-making, H-GEM ensures differentiated expert specialization and cross-task modeling. Experiments on CMU-MOSI and CMU-MOSEI datasets demonstrate that H-GEM outperforms baseline models across a series of metrics. Compared with single-layer MoE architectures, the significantly reduced routing entropy indicates effective mitigation of expert redundancy. Moreover, the proposed model demonstrates higher robustness in low-resource and modality-missing scenarios, highlighting its strong practical applicability.
Available online: January 19, 2026
,DOI: 10.15888/j.cnki.csa.010110
Abstract:
Existing multimodal fake news detection methods still suffer from the following limitations. During cross-modal semantic alignment, only global features are typically aligned, failing to establish fine-grained semantic correspondences between local image regions and their relevant text fragments. In the modality fusion stage, an equal-weight combination strategy is usually adopted, which prevents the more informative dominant modality from being fully utilized, thereby limiting model performance. To address these issues, this study proposes a multimodal fake news detection model integrating fine-grained alignment and dominant modality enhancement. The fine-grained alignment module leverages the capability of the FG-CLIP model to guide precise correspondence between deep semantic features of news images and texts, effectively suppressing interference from irrelevant regions. Moreover, the dominant modality is determined by a confidence score, which is computed based on the distance between single-modality features and their corresponding class prototypes. A prototype cross-entropy loss is introduced to enhance the representational capacity of the dominant modality, enabling it to play a leading role in the fusion process. Experimental results on the Weibo and GossipCop datasets demonstrate that the proposed model outperforms baseline methods on most evaluation metrics, verifying its effectiveness and robustness in fake news detection tasks.
Existing multimodal fake news detection methods still suffer from the following limitations. During cross-modal semantic alignment, only global features are typically aligned, failing to establish fine-grained semantic correspondences between local image regions and their relevant text fragments. In the modality fusion stage, an equal-weight combination strategy is usually adopted, which prevents the more informative dominant modality from being fully utilized, thereby limiting model performance. To address these issues, this study proposes a multimodal fake news detection model integrating fine-grained alignment and dominant modality enhancement. The fine-grained alignment module leverages the capability of the FG-CLIP model to guide precise correspondence between deep semantic features of news images and texts, effectively suppressing interference from irrelevant regions. Moreover, the dominant modality is determined by a confidence score, which is computed based on the distance between single-modality features and their corresponding class prototypes. A prototype cross-entropy loss is introduced to enhance the representational capacity of the dominant modality, enabling it to play a leading role in the fusion process. Experimental results on the Weibo and GossipCop datasets demonstrate that the proposed model outperforms baseline methods on most evaluation metrics, verifying its effectiveness and robustness in fake news detection tasks.
Available online: January 16, 2026
,DOI: 10.15888/j.cnki.csa.010112
Abstract:
Significant radiometric and geometric discrepancies among multimodal remote sensing images present substantial challenges for achieving high-precision registration. To address these issues, this study proposes a phase-congruency-enhanced adjacent self-similarity matching method, termed as PC-ASS, for multimodal remote sensing imagery. First, a multi-scale image representation is constructed via nonlinear diffusion filtering to suppress noise while preserving common edges and structural information, thus providing a reliable foundation for subsequent feature detection. Next, phase congruency amplitude maps are computed using multi-scale, multi-orientation Log-Gabor filters to characterize structurally salient regions in the images. The phase congruency amplitudes are then used as weighting factors in computing adjacent self-similarity responses, thus enhancing structural features: regions with higher phase congruency yield stronger responses, increasing both the number and the quality of robust features such as edges and corners. Furthermore, during descriptor construction, a phase-congruency-weighted mechanism is incorporated into the polar statistical histogram framework, weighting each pixel’s adjacent self-similarity value by its phase congruency amplitude. This ensures that structurally salient regions contribute more prominently to the descriptor, thereby improving robustness against noise, texture interference, and cross-modal radiometric differences. Finally, incorrect matches are eliminated through a nearest-neighbor distance ratio strategy combined with the fast sample consensus (FSC) algorithm, enabling high-precision registration. Comparative experiments on three publicly available multimodal remote sensing datasets against five representative methods (PSO-SIFT, OSS, HAPCG, RIFT, and ASS) demonstrate that PC-ASS outperforms existing approaches in average correct matches, mean root-mean-square error, and correct matching rate, highlighting its robustness and broad applicability.
Significant radiometric and geometric discrepancies among multimodal remote sensing images present substantial challenges for achieving high-precision registration. To address these issues, this study proposes a phase-congruency-enhanced adjacent self-similarity matching method, termed as PC-ASS, for multimodal remote sensing imagery. First, a multi-scale image representation is constructed via nonlinear diffusion filtering to suppress noise while preserving common edges and structural information, thus providing a reliable foundation for subsequent feature detection. Next, phase congruency amplitude maps are computed using multi-scale, multi-orientation Log-Gabor filters to characterize structurally salient regions in the images. The phase congruency amplitudes are then used as weighting factors in computing adjacent self-similarity responses, thus enhancing structural features: regions with higher phase congruency yield stronger responses, increasing both the number and the quality of robust features such as edges and corners. Furthermore, during descriptor construction, a phase-congruency-weighted mechanism is incorporated into the polar statistical histogram framework, weighting each pixel’s adjacent self-similarity value by its phase congruency amplitude. This ensures that structurally salient regions contribute more prominently to the descriptor, thereby improving robustness against noise, texture interference, and cross-modal radiometric differences. Finally, incorrect matches are eliminated through a nearest-neighbor distance ratio strategy combined with the fast sample consensus (FSC) algorithm, enabling high-precision registration. Comparative experiments on three publicly available multimodal remote sensing datasets against five representative methods (PSO-SIFT, OSS, HAPCG, RIFT, and ASS) demonstrate that PC-ASS outperforms existing approaches in average correct matches, mean root-mean-square error, and correct matching rate, highlighting its robustness and broad applicability.
Available online: January 16, 2026
,DOI: 10.15888/j.cnki.csa.010109
Abstract:
To address the insufficient generalization and degraded detection performance in intrusion detection models caused by scarce samples during the early stages of novel network attacks, this study proposes a few-shot intrusion detection method based on diffusion models. At the data augmentation level, the proposed method introduces a noise-aware conditional diffusion model that employs cosine noise scheduling to balance generation efficiency and sample fidelity, while residual connections are incorporated to enhance feature propagation stability and improve the distribution fidelity of synthesized traffic data. At the feature metric level, a dynamic prototype network is designed, leveraging multi-head attention to optimize class prototype representations and mitigate feature sparsity in few-shot scenarios. Simultaneously, a joint optimization strategy combining cross-entropy loss with an orthogonal regularization term is adopted to enhance intra-class compactness and increase inter-class separability. Experimental results on two public datasets demonstrate that the proposed model outperforms other detection methods in terms of accuracy and generalization capability under few-shot scenarios, providing a novel solution approach for few-shot intrusion detection.
To address the insufficient generalization and degraded detection performance in intrusion detection models caused by scarce samples during the early stages of novel network attacks, this study proposes a few-shot intrusion detection method based on diffusion models. At the data augmentation level, the proposed method introduces a noise-aware conditional diffusion model that employs cosine noise scheduling to balance generation efficiency and sample fidelity, while residual connections are incorporated to enhance feature propagation stability and improve the distribution fidelity of synthesized traffic data. At the feature metric level, a dynamic prototype network is designed, leveraging multi-head attention to optimize class prototype representations and mitigate feature sparsity in few-shot scenarios. Simultaneously, a joint optimization strategy combining cross-entropy loss with an orthogonal regularization term is adopted to enhance intra-class compactness and increase inter-class separability. Experimental results on two public datasets demonstrate that the proposed model outperforms other detection methods in terms of accuracy and generalization capability under few-shot scenarios, providing a novel solution approach for few-shot intrusion detection.
Available online: January 15, 2026
,DOI: 10.15888/j.cnki.csa.010101
Abstract:
A reinforcement-learning-based multi-strategy Harris hawks optimization (RLMHHO) is developed to address the scheduling problem of distributed hybrid flow-shop, with the optimization objectives of minimizing the maximum completion time and delay time. The algorithm uses a grouping chaos initialization strategy to improve the randomness and diversity of the initial search. A four-group eagle management mechanism of exploration, development, balance, and elite is introduced to achieve synergy between global search and local development. A reinforcement learning coordinator based on deep Q-networks dynamically selects the optimal search strategy based on a 14-dimensional state space. Simulation experiments have verified that the proposed algorithm offers better solution quality and stronger search capability for solving this type of scheduling problem.
A reinforcement-learning-based multi-strategy Harris hawks optimization (RLMHHO) is developed to address the scheduling problem of distributed hybrid flow-shop, with the optimization objectives of minimizing the maximum completion time and delay time. The algorithm uses a grouping chaos initialization strategy to improve the randomness and diversity of the initial search. A four-group eagle management mechanism of exploration, development, balance, and elite is introduced to achieve synergy between global search and local development. A reinforcement learning coordinator based on deep Q-networks dynamically selects the optimal search strategy based on a 14-dimensional state space. Simulation experiments have verified that the proposed algorithm offers better solution quality and stronger search capability for solving this type of scheduling problem.
Available online: January 15, 2026
,DOI: 10.15888/j.cnki.csa.010108
Abstract:
Communication signal generation is fundamental for designing and optimizing intelligent communication systems under few-shot conditions. Existing methods perform well in static environments but struggle to capture rapidly evolving spatio-temporal features in dynamic scenarios, leading to reduced consistency and accuracy. To address this, this study proposes a Transformer-based diffusion model, named TransDiffusion, for high-dynamic communication signal generation. TransDiffusion integrates Transformer architecture into the diffusion framework and enhances long-range dependency modeling through embedded multi-level spatio-temporal attention. The noise prediction network is also optimized for time-varying environmental characteristics. Experiments on a custom high-dynamic simulation dataset that emulates multi-target motion and RF signatures in urban traffic show that TransDiffusion significantly outperforms the RF-Diffusion baseline: MME, MSE, and MAE are reduced by 85.13%, 40.92%, and 30.62%, while STFT-sim and Spectral-sim increase by 154.30% and 5.28%, respectively. This demonstrates the effectiveness of the proposed method in reconstructing high-fidelity communication signals in dynamic scenarios.
Communication signal generation is fundamental for designing and optimizing intelligent communication systems under few-shot conditions. Existing methods perform well in static environments but struggle to capture rapidly evolving spatio-temporal features in dynamic scenarios, leading to reduced consistency and accuracy. To address this, this study proposes a Transformer-based diffusion model, named TransDiffusion, for high-dynamic communication signal generation. TransDiffusion integrates Transformer architecture into the diffusion framework and enhances long-range dependency modeling through embedded multi-level spatio-temporal attention. The noise prediction network is also optimized for time-varying environmental characteristics. Experiments on a custom high-dynamic simulation dataset that emulates multi-target motion and RF signatures in urban traffic show that TransDiffusion significantly outperforms the RF-Diffusion baseline: MME, MSE, and MAE are reduced by 85.13%, 40.92%, and 30.62%, while STFT-sim and Spectral-sim increase by 154.30% and 5.28%, respectively. This demonstrates the effectiveness of the proposed method in reconstructing high-fidelity communication signals in dynamic scenarios.
Available online: January 15, 2026
,DOI: 10.15888/j.cnki.csa.010104
Abstract:
With the deep integration of the Internet of Vehicles (IoV) and mobile edge computing (MEC), achieving efficient task offloading while ensuring low latency and high reliability has become a key challenge. Although the Harris hawks optimization (HHO) algorithm demonstrates strong global search capability, it still suffers from limitations in population initialization, exploration–exploitation transition, and diversity maintenance, making it prone to premature convergence. To address these issues, this study proposes a dynamic dual-population Harris hawks optimization (DDHHO) algorithm. The proposed algorithm introduces a dynamic dual-population co-evolution (DDPC) mechanism, combined with L-T chaotic initialization, nonlinear escape energy, and a nonlinear jump strategy, to adaptively balance global exploration and local exploitation. Experimental results show that in the IoV-MEC task offloading scenario, DDHHO reduces the total system cost by approximately 2.5%, 3.2%, 4.9%, 6.0%, and 7.9% compared with the mixed-strategy HHO (MSHHO), original HHO, MASSFOA, IPSO and PSO, respectively. Moreover, DDHHO exhibits faster convergence speed and higher stability in joint latency-energy optimization. These results verify the effectiveness and superiority of DDHHO, providing an efficient, stable, and scalable optimization solution for resource management in IoV-MEC systems.
With the deep integration of the Internet of Vehicles (IoV) and mobile edge computing (MEC), achieving efficient task offloading while ensuring low latency and high reliability has become a key challenge. Although the Harris hawks optimization (HHO) algorithm demonstrates strong global search capability, it still suffers from limitations in population initialization, exploration–exploitation transition, and diversity maintenance, making it prone to premature convergence. To address these issues, this study proposes a dynamic dual-population Harris hawks optimization (DDHHO) algorithm. The proposed algorithm introduces a dynamic dual-population co-evolution (DDPC) mechanism, combined with L-T chaotic initialization, nonlinear escape energy, and a nonlinear jump strategy, to adaptively balance global exploration and local exploitation. Experimental results show that in the IoV-MEC task offloading scenario, DDHHO reduces the total system cost by approximately 2.5%, 3.2%, 4.9%, 6.0%, and 7.9% compared with the mixed-strategy HHO (MSHHO), original HHO, MASSFOA, IPSO and PSO, respectively. Moreover, DDHHO exhibits faster convergence speed and higher stability in joint latency-energy optimization. These results verify the effectiveness and superiority of DDHHO, providing an efficient, stable, and scalable optimization solution for resource management in IoV-MEC systems.
Available online: January 15, 2026
,DOI: 10.15888/j.cnki.csa.010113
Abstract:
With the increasing frequency of extreme climate events, enhancing precipitation forecasting capability has become an urgent need in meteorological operations. Most existing data-driven methods model precipitation motion and intensity in a coupled manner. Although effective at capturing large-scale precipitation systems, these methods struggle to accurately predict the rapid evolution of small- to medium-scale heavy precipitation, thereby limiting their forecast skill for intense rainfall events. In this study, a dual-branch fusion network named LAFUNet is proposed, which decouples motion and intensity via Lagrangian transformation. One branch directly analyzes original radar image sequences to capture the spatial structure and motion characteristics of large-scale precipitation systems. The other branch transforms the precipitation field into the Lagrangian coordinates, focusing on modeling intensity evolution to better represent the nonlinear intensity changes associated with small- to medium-scale heavy precipitation. Additionally, a dual-branch interaction module is designed to adaptively fuse features from both branches. Experiments are conducted on the public CIKM and SEVIR radar datasets. The results demonstrate that the proposed model achieves outstanding performance in heavy precipitation nowcasting. Particularly on the SEVIR dataset, for extreme precipitation events with an intensity threshold exceeding 219, the model attains a CSI score of 0.1368 for 1 hour forecasts, significantly outperforming comparative models such as VMRNN.
With the increasing frequency of extreme climate events, enhancing precipitation forecasting capability has become an urgent need in meteorological operations. Most existing data-driven methods model precipitation motion and intensity in a coupled manner. Although effective at capturing large-scale precipitation systems, these methods struggle to accurately predict the rapid evolution of small- to medium-scale heavy precipitation, thereby limiting their forecast skill for intense rainfall events. In this study, a dual-branch fusion network named LAFUNet is proposed, which decouples motion and intensity via Lagrangian transformation. One branch directly analyzes original radar image sequences to capture the spatial structure and motion characteristics of large-scale precipitation systems. The other branch transforms the precipitation field into the Lagrangian coordinates, focusing on modeling intensity evolution to better represent the nonlinear intensity changes associated with small- to medium-scale heavy precipitation. Additionally, a dual-branch interaction module is designed to adaptively fuse features from both branches. Experiments are conducted on the public CIKM and SEVIR radar datasets. The results demonstrate that the proposed model achieves outstanding performance in heavy precipitation nowcasting. Particularly on the SEVIR dataset, for extreme precipitation events with an intensity threshold exceeding 219, the model attains a CSI score of 0.1368 for 1 hour forecasts, significantly outperforming comparative models such as VMRNN.
Available online: January 15, 2026
,DOI: 10.15888/j.cnki.csa.010099
Abstract:
Due to its highly destructive nature and significant disaster risk, accurate lightning monitoring is a crucial component of disaster prevention and mitigation. Geostationary meteorological satellites, such as Himawari-8, provide an ideal platform for lightning monitoring with the advantages of broad coverage and continuous observation. However, the physical correlation mechanism between satellite cloud images and lightning activity remains unclear, which restricts practical application. This study proposes the AE-UNet model, which utilizes Himawari satellite data and VLF long-range lightning detection network data, identifying high-density lightning areas from satellite cloud images. The AE-UNet model incorporates a channel attention mechanism and residual connections to adaptively and deeply fuse the multi-channel satellite features. During the concatenation of multi-scale features, a channel-space dual attention mechanism is embedded to fully explore spatial correlations. The experimental results show that the AE-UNet model achieves a lightning identification accuracy of 97.91%, a probability of detection (POD) of 67.47%, and a false alarm rate (FAR) of 26.92%, demonstrating significant performance improvement over the benchmark model. The proposed model can provide reliable lightning activity information based on satellite cloud images, thereby strongly supporting disaster prevention and mitigation efforts.
Due to its highly destructive nature and significant disaster risk, accurate lightning monitoring is a crucial component of disaster prevention and mitigation. Geostationary meteorological satellites, such as Himawari-8, provide an ideal platform for lightning monitoring with the advantages of broad coverage and continuous observation. However, the physical correlation mechanism between satellite cloud images and lightning activity remains unclear, which restricts practical application. This study proposes the AE-UNet model, which utilizes Himawari satellite data and VLF long-range lightning detection network data, identifying high-density lightning areas from satellite cloud images. The AE-UNet model incorporates a channel attention mechanism and residual connections to adaptively and deeply fuse the multi-channel satellite features. During the concatenation of multi-scale features, a channel-space dual attention mechanism is embedded to fully explore spatial correlations. The experimental results show that the AE-UNet model achieves a lightning identification accuracy of 97.91%, a probability of detection (POD) of 67.47%, and a false alarm rate (FAR) of 26.92%, demonstrating significant performance improvement over the benchmark model. The proposed model can provide reliable lightning activity information based on satellite cloud images, thereby strongly supporting disaster prevention and mitigation efforts.
Available online: January 15, 2026
,DOI: 10.15888/j.cnki.csa.010100
Abstract:
Abstractive text summarization aims to generate concise and readable summaries by understanding the original input text. However, the summaries produced by existing models still face issues such as semantic redundancy, factual errors, and exposure bias. Addressing these problems is crucial for improving model performance and summary quality. Therefore, an abstractive text summarization model that integrates knowledge enhancement with the SimCLS framework is proposed. First, a knowledge-enhanced encoder is designed to obtain structured knowledge information from the source text to preserve the structural information of the global context, and it is combined with a text encoder to fully capture the semantic information of the entire text. Then, the copy mechanism is utilized in the decoder to more accurately reproduce the information from the original text. Finally, the summaries generated by the model are scored using the SimCLS two-stage contrastive learning framework to guide the generation of high-quality summaries. The experimental results show that, when compared with the higher-performing SeqCo model, the proposed model improves ROUGE-1/2/L and BERTScore on the CNN/Daily Mail dataset by 1.84, 0.65, 2.04, and 0.21 percentage points, respectively, and on the XSum dataset by 1.78, 2.16, 2.36, and 0.13 percentage points, confirming the model’s effectiveness.
Abstractive text summarization aims to generate concise and readable summaries by understanding the original input text. However, the summaries produced by existing models still face issues such as semantic redundancy, factual errors, and exposure bias. Addressing these problems is crucial for improving model performance and summary quality. Therefore, an abstractive text summarization model that integrates knowledge enhancement with the SimCLS framework is proposed. First, a knowledge-enhanced encoder is designed to obtain structured knowledge information from the source text to preserve the structural information of the global context, and it is combined with a text encoder to fully capture the semantic information of the entire text. Then, the copy mechanism is utilized in the decoder to more accurately reproduce the information from the original text. Finally, the summaries generated by the model are scored using the SimCLS two-stage contrastive learning framework to guide the generation of high-quality summaries. The experimental results show that, when compared with the higher-performing SeqCo model, the proposed model improves ROUGE-1/2/L and BERTScore on the CNN/Daily Mail dataset by 1.84, 0.65, 2.04, and 0.21 percentage points, respectively, and on the XSum dataset by 1.78, 2.16, 2.36, and 0.13 percentage points, confirming the model’s effectiveness.
Available online: January 15, 2026
,DOI: 10.15888/j.cnki.csa.010096
Abstract:
The transmission line channels feature complex environments, and various hidden external damage targets exhibit significant scale differences due to factors such as shooting angles and observation distances, thereby resulting in the model’s low precision and prominent problems of false and missed detection in diversified risk target recognition. To this end, this study proposes a transmission line external damage detection method based on hierarchical feature fusion. The method is based on the RT-DETR model and introduces a lightweight C2f_MambaOut module to optimize the backbone structure and effectively reduce model parameters. Additionally, a PA_CGLU module integrating polarity-aware attention and gating mechanisms is established to replace the original AIFI module, thereby enhancing the query vectors’ directional perception of image features and salient modeling capabilities, as well as improving adaptive semantic matching efficiency. Furthermore, a hierarchical attention fusion block (HAFB) is designed to realize multi-scale hierarchical fusion and enhancement of input features by employing local and global attention branches, thus boosting the comprehensive recognition ability of multi-category and multi-scale targets. Additionally, a transmission line external damage detection dataset that covers various real-world scenarios and features a balanced distribution of samples is constructed. Experimental results on this dataset demonstrate that the improved model achieves a 1.5% increase in mean average precision (mAP) and a 20.7% reduction in the parameter count. The results demonstrate that the effectiveness of the proposed method in mitigating the challenges posed by target scale variation and enhancing the overall detection performance for diverse external damage risks, thereby achieving a better balance between model efficiency and accuracy.
The transmission line channels feature complex environments, and various hidden external damage targets exhibit significant scale differences due to factors such as shooting angles and observation distances, thereby resulting in the model’s low precision and prominent problems of false and missed detection in diversified risk target recognition. To this end, this study proposes a transmission line external damage detection method based on hierarchical feature fusion. The method is based on the RT-DETR model and introduces a lightweight C2f_MambaOut module to optimize the backbone structure and effectively reduce model parameters. Additionally, a PA_CGLU module integrating polarity-aware attention and gating mechanisms is established to replace the original AIFI module, thereby enhancing the query vectors’ directional perception of image features and salient modeling capabilities, as well as improving adaptive semantic matching efficiency. Furthermore, a hierarchical attention fusion block (HAFB) is designed to realize multi-scale hierarchical fusion and enhancement of input features by employing local and global attention branches, thus boosting the comprehensive recognition ability of multi-category and multi-scale targets. Additionally, a transmission line external damage detection dataset that covers various real-world scenarios and features a balanced distribution of samples is constructed. Experimental results on this dataset demonstrate that the improved model achieves a 1.5% increase in mean average precision (mAP) and a 20.7% reduction in the parameter count. The results demonstrate that the effectiveness of the proposed method in mitigating the challenges posed by target scale variation and enhancing the overall detection performance for diverse external damage risks, thereby achieving a better balance between model efficiency and accuracy.
Available online: January 08, 2026
,DOI: 10.15888/j.cnki.csa.010097
Abstract:
In the field of electromagnetic information security, the detection of electromagnetic leakage of red signals is severely affected by electromagnetic interference. Traditional denoising methods exhibit limitations when dealing with non-stationary signals and complex noise environments. In this study, a denoising method based on generative adversarial network (GAN) is proposed, and efficient noise reduction is achieved through adversarial learning between the generator and discriminator. To address the non-stationary characteristics of electromagnetic signals, a time-frequency dual-domain attention mechanism (TF-DAM) is designed. The generator adopts an improved U-Net architecture that incorporates TF-DAM and integrates residual networks and dropout layers to enhance generalization capability. The encoder-decoder structure and skip connections are utilized to preserve signal details. During training, a dynamic loss-weight adjustment strategy is employed to improve training efficiency and denoising performance. Experimental results demonstrate that the proposed method outperforms traditional approaches in terms of signal-to-noise ratio (SNR) improvement and detail preservation, exhibiting superior performance in nonstationary signal processing. This study provides a novel solution for electromagnetic signal denoising, demonstrating high practical application value.
In the field of electromagnetic information security, the detection of electromagnetic leakage of red signals is severely affected by electromagnetic interference. Traditional denoising methods exhibit limitations when dealing with non-stationary signals and complex noise environments. In this study, a denoising method based on generative adversarial network (GAN) is proposed, and efficient noise reduction is achieved through adversarial learning between the generator and discriminator. To address the non-stationary characteristics of electromagnetic signals, a time-frequency dual-domain attention mechanism (TF-DAM) is designed. The generator adopts an improved U-Net architecture that incorporates TF-DAM and integrates residual networks and dropout layers to enhance generalization capability. The encoder-decoder structure and skip connections are utilized to preserve signal details. During training, a dynamic loss-weight adjustment strategy is employed to improve training efficiency and denoising performance. Experimental results demonstrate that the proposed method outperforms traditional approaches in terms of signal-to-noise ratio (SNR) improvement and detail preservation, exhibiting superior performance in nonstationary signal processing. This study provides a novel solution for electromagnetic signal denoising, demonstrating high practical application value.
Available online: January 08, 2026
,DOI: 10.15888/j.cnki.csa.010089
Abstract:
Prostate cancer is a common malignant tumor among men, and the automatic anomaly detection of its MRI images is crucial for improving diagnostic efficiency and reducing doctors’ workload. To address the problem that existing methods rely on large amounts of labeled data and have limited ability to detect subtle lesions, this paper proposes an unsupervised prostate cancer anomaly detection method based on ensemble autoencoders (EAE) and residual-based pseudo-anomaly generation. Specifically, an EAE with centered kernel alignment (CKA) loss is first introduced to suppress feature redundancy and produce diverse reconstruction results. Secondly, by integrating the consensus high-response regions from the multi-branch EAE, anatomically constrained perturbations are injected in the condition of no manual labels to generate pseudo-anomaly samples that more closely resemble real lesions, all without manual labels. Furthermore, a frequency-domain contrastive loss is introduced in the detection network to enhance the ability to distinguish between normal and anomaly samples in the frequency domain space. Finally, a two-stage process is adopted to maintain consistency between training and inference, thus improving model robustness. Extensive experiments show that the proposed method yields superior detection accuracy and generalization ability on public prostate MRI and brain MRI datasets. Specifically, on datasets such as PICAI, PMU and BTM, the image-level AUC reaches 84.00%, 89.20%, and 89.50% respectively, and the AP values are 82.5%, 87.00%, and 88.80% respectively, outperforming current state-of-the-art methods.
Prostate cancer is a common malignant tumor among men, and the automatic anomaly detection of its MRI images is crucial for improving diagnostic efficiency and reducing doctors’ workload. To address the problem that existing methods rely on large amounts of labeled data and have limited ability to detect subtle lesions, this paper proposes an unsupervised prostate cancer anomaly detection method based on ensemble autoencoders (EAE) and residual-based pseudo-anomaly generation. Specifically, an EAE with centered kernel alignment (CKA) loss is first introduced to suppress feature redundancy and produce diverse reconstruction results. Secondly, by integrating the consensus high-response regions from the multi-branch EAE, anatomically constrained perturbations are injected in the condition of no manual labels to generate pseudo-anomaly samples that more closely resemble real lesions, all without manual labels. Furthermore, a frequency-domain contrastive loss is introduced in the detection network to enhance the ability to distinguish between normal and anomaly samples in the frequency domain space. Finally, a two-stage process is adopted to maintain consistency between training and inference, thus improving model robustness. Extensive experiments show that the proposed method yields superior detection accuracy and generalization ability on public prostate MRI and brain MRI datasets. Specifically, on datasets such as PICAI, PMU and BTM, the image-level AUC reaches 84.00%, 89.20%, and 89.50% respectively, and the AP values are 82.5%, 87.00%, and 88.80% respectively, outperforming current state-of-the-art methods.
Available online: January 08, 2026
,DOI: 10.15888/j.cnki.csa.010095
Abstract:
The current research on KL divergence-based fuzzy clustering segmentation faces two core challenges. The first is how to effectively balance the noise tolerance and computational efficiency of the algorithms to meet the requirements for real-time performance. The second is how to avoid the local optimum caused by non-convex objective functions and improve the accuracy and stability of complex images. To this end, this study proposes a fast fuzzy clustering image segmentation algorithm that integrates intra-class and inter-class distances and KL divergence. Firstly, the classical idea of solely minimizing the intra-class distance is abandoned, and a new objective metric based on the difference of minimizing the intra-class distance and maximizing the inter-class distance is constructed, which ensures the minimization of intra-class distance and maximization of inter-class distance. Additionally, this guarantees that the sample points can accurately find its corresponding class during classification. Secondly, KL divergence is innovatively combined with histograms of images. On the one hand, KL divergence is employed to enhance the robustness for noise and non-uniform data. On the other hand, the utilization of the histograms can greatly reduce the data amount of algorithm iteration, which improves the local consistency and ensures the algorithms’ efficiency. As a result, the difficulty of existing methods in balancing “robustness”, “accuracy” and “real-time performance” is solved to make the algorithm more applicable to fields such as medicine, intelligent driving, and robot navigation. A large number of various kinds of image segmentation tests show that the proposed intra-class and inter-class KL divergence-based fuzzy C-means clustering algorithm is effective, especially for segmenting large images with big noise. The algorithm can not only remove noises but also satisfy the requirements of real-time segmentation.
The current research on KL divergence-based fuzzy clustering segmentation faces two core challenges. The first is how to effectively balance the noise tolerance and computational efficiency of the algorithms to meet the requirements for real-time performance. The second is how to avoid the local optimum caused by non-convex objective functions and improve the accuracy and stability of complex images. To this end, this study proposes a fast fuzzy clustering image segmentation algorithm that integrates intra-class and inter-class distances and KL divergence. Firstly, the classical idea of solely minimizing the intra-class distance is abandoned, and a new objective metric based on the difference of minimizing the intra-class distance and maximizing the inter-class distance is constructed, which ensures the minimization of intra-class distance and maximization of inter-class distance. Additionally, this guarantees that the sample points can accurately find its corresponding class during classification. Secondly, KL divergence is innovatively combined with histograms of images. On the one hand, KL divergence is employed to enhance the robustness for noise and non-uniform data. On the other hand, the utilization of the histograms can greatly reduce the data amount of algorithm iteration, which improves the local consistency and ensures the algorithms’ efficiency. As a result, the difficulty of existing methods in balancing “robustness”, “accuracy” and “real-time performance” is solved to make the algorithm more applicable to fields such as medicine, intelligent driving, and robot navigation. A large number of various kinds of image segmentation tests show that the proposed intra-class and inter-class KL divergence-based fuzzy C-means clustering algorithm is effective, especially for segmenting large images with big noise. The algorithm can not only remove noises but also satisfy the requirements of real-time segmentation.
Available online: January 08, 2026
,DOI: 10.15888/j.cnki.csa.010091
Abstract:
This study proposes an efficient Mamba-Transformer collaborated network, termed MTSR, to address the challenges of balancing the global receptive field, local feature extraction, and computational efficiency in lightweight image super-resolution models, while tackling the insufficient cross-token interaction capability in Mamba-based context modeling. First, a hybrid synergy architecture is constructed by integrating Mamba and Transformer modules in an optimal ratio. This design leverages the Transformer’s superior cross-token interaction ability to compensate for the contextual modeling deficiencies in purely Mamba-based models. Second, a novel deep convolutional attention feed-forward network is designed to replace the traditional multi-layer perceptron. This network significantly enhances the capability for local detail extraction and effective inter-channel communication, thus reducing pixel-level information loss and amplifying the performance potential of the Mamba modules. Finally, a triple depthwise-separable shallow refinement block is introduced to efficiently capture and enhance the shallow-level features of an image, thus providing richer original texture information for subsequent nonlinear mapping. Extensive experiments on five public benchmark datasets demonstrate that the proposed MTSR model achieves peak signal-to-noise ratio (PSNR) gains of up to 0.31 and 0.38 dB over the state-of-the-art lightweight models SRFormer-light and MambaIR-light, respectively, while maintaining a highly competitive inference speed. This study validates that the proposed method provides an effective solution for the field of lightweight image super-resolution, combining both high performance and high efficiency.
This study proposes an efficient Mamba-Transformer collaborated network, termed MTSR, to address the challenges of balancing the global receptive field, local feature extraction, and computational efficiency in lightweight image super-resolution models, while tackling the insufficient cross-token interaction capability in Mamba-based context modeling. First, a hybrid synergy architecture is constructed by integrating Mamba and Transformer modules in an optimal ratio. This design leverages the Transformer’s superior cross-token interaction ability to compensate for the contextual modeling deficiencies in purely Mamba-based models. Second, a novel deep convolutional attention feed-forward network is designed to replace the traditional multi-layer perceptron. This network significantly enhances the capability for local detail extraction and effective inter-channel communication, thus reducing pixel-level information loss and amplifying the performance potential of the Mamba modules. Finally, a triple depthwise-separable shallow refinement block is introduced to efficiently capture and enhance the shallow-level features of an image, thus providing richer original texture information for subsequent nonlinear mapping. Extensive experiments on five public benchmark datasets demonstrate that the proposed MTSR model achieves peak signal-to-noise ratio (PSNR) gains of up to 0.31 and 0.38 dB over the state-of-the-art lightweight models SRFormer-light and MambaIR-light, respectively, while maintaining a highly competitive inference speed. This study validates that the proposed method provides an effective solution for the field of lightweight image super-resolution, combining both high performance and high efficiency.
Available online: January 08, 2026
,DOI: 10.15888/j.cnki.csa.010084
Abstract:
Federated learning is widely applied in mobile networks, utilizing a distributed approach for both local and global model training, effectively ensuring data security. However, its application in Internet of Vehicles faces three major challenges: frequent global communication causing high latency, limited computational and power resources of vehicles, and bandwidth fluctuations induced by highly dynamic network environments. These factors collectively hinder the execution of federated learning systems and significantly reduce efficiency. To address these issues, this study proposes an adaptive resource scheduling strategy integrating mobility and bandwidth prediction with multi-agent reinforcement learning. This strategy introduces a multidimensional prediction model combining wavelet neural network (WNN) and extended long short-term memory (xLSTM) to forecast vehicle mobility positions and bandwidth, providing accurate inputs for the multi-agent reinforcement learning framework. Concurrently, a multi-agent deep Q-network (MADQN) dynamic resource scheduling algorithm is employed to tackle resource allocation and power control challenges, thus mitigating the impact of the tail problem on system efficiency and enabling autonomous decision-making by edge nodes. Experimental results show that, compared to traditional methods, this approach significantly reduces system costs and energy consumption, while improving model performance and transmission success rates.
Federated learning is widely applied in mobile networks, utilizing a distributed approach for both local and global model training, effectively ensuring data security. However, its application in Internet of Vehicles faces three major challenges: frequent global communication causing high latency, limited computational and power resources of vehicles, and bandwidth fluctuations induced by highly dynamic network environments. These factors collectively hinder the execution of federated learning systems and significantly reduce efficiency. To address these issues, this study proposes an adaptive resource scheduling strategy integrating mobility and bandwidth prediction with multi-agent reinforcement learning. This strategy introduces a multidimensional prediction model combining wavelet neural network (WNN) and extended long short-term memory (xLSTM) to forecast vehicle mobility positions and bandwidth, providing accurate inputs for the multi-agent reinforcement learning framework. Concurrently, a multi-agent deep Q-network (MADQN) dynamic resource scheduling algorithm is employed to tackle resource allocation and power control challenges, thus mitigating the impact of the tail problem on system efficiency and enabling autonomous decision-making by edge nodes. Experimental results show that, compared to traditional methods, this approach significantly reduces system costs and energy consumption, while improving model performance and transmission success rates.
Available online: January 08, 2026
,DOI: 10.15888/j.cnki.csa.010085
Abstract:
To address the issues of insufficient detection accuracy and low model inference efficiency in wrist fracture detection for medical image analysis in emergency scenarios, this study proposes a lightweight wrist fracture detection algorithm based on the improved YOLO11, named MDM-YOLO. First, a multi-scale feature extraction module is designed to address the multi-scale representation challenges of complex fracture shapes by extracting information at different scales through multiple parallel branches. Second, a mixed spatial local attention mechanism is proposed, which combines local and global features to significantly enhance the attention to subtle fractures. Finally, a dynamic depthwise separable convolution (DDSConv) is designed to reduce computational complexity and accelerate inference speed while maintaining detection accuracy, thus making the model more lightweight. Experiments show that MDM-YOLO achieves an precision of 92.6%, a recall rate of 88.1%, and an mAP50 of 95.1% on the GRAZPEDWRI-DX dataset, representing improvements of 1.7%, 2.5%, and 1.5%, respectively, compared with the original model. Under the same hardware conditions, the detection speed increases by 37%, and the number of parameters is only 73.3% of the original model, verifying the effectiveness of the lightweight design. This provides an efficient solution for rapid wrist fracture diagnosis in emergency scenarios.
To address the issues of insufficient detection accuracy and low model inference efficiency in wrist fracture detection for medical image analysis in emergency scenarios, this study proposes a lightweight wrist fracture detection algorithm based on the improved YOLO11, named MDM-YOLO. First, a multi-scale feature extraction module is designed to address the multi-scale representation challenges of complex fracture shapes by extracting information at different scales through multiple parallel branches. Second, a mixed spatial local attention mechanism is proposed, which combines local and global features to significantly enhance the attention to subtle fractures. Finally, a dynamic depthwise separable convolution (DDSConv) is designed to reduce computational complexity and accelerate inference speed while maintaining detection accuracy, thus making the model more lightweight. Experiments show that MDM-YOLO achieves an precision of 92.6%, a recall rate of 88.1%, and an mAP50 of 95.1% on the GRAZPEDWRI-DX dataset, representing improvements of 1.7%, 2.5%, and 1.5%, respectively, compared with the original model. Under the same hardware conditions, the detection speed increases by 37%, and the number of parameters is only 73.3% of the original model, verifying the effectiveness of the lightweight design. This provides an efficient solution for rapid wrist fracture diagnosis in emergency scenarios.
Available online: January 08, 2026
,DOI: 10.15888/j.cnki.csa.010086
Abstract:
To address the inefficiencies, high costs, and safety risks inherent in traditional road and bridge inspection techniques, as well as the challenges posed by the large parameter volumes of current multimodal detection models and the difficulty in achieving real-time deployment on unmanned aerial vehicle (UAV) platforms, this study proposes a multimodal feature fusion road and bridge detection model based on cross distillation. The model employs a dual-branch teacher network and a single-branch student network architecture. Efficient knowledge transfer of modality-specific features is achieved through feature interaction and collaborative distillation mechanisms between the teacher networks. Concurrently, a dynamic feature fusion module, utilizing attention mechanisms, is introduced to enhance the perception of critical features associated with road and bridge defects. Experimental results demonstrate that, while maintaining a detection precision of 89.6% mAP@0.5, the proposed model reduces its parameter size to 8.2M and achieves an inference speed of 32.6 f/s. These results significantly outperform traditional multimodal fusion and lightweight methods. Compared to strategies utilizing feature concatenation or post-distillation unimodal fusion, the proposed model shows clear advantages in both detection accuracy and computational efficiency. Ablation studies confirm the effectiveness of the cross-distillation mechanism and the attention-based fusion module. The model successfully enables high-precision, lightweight detection of road and bridge defects, thus providing a technical foundation for the engineering application of UAV-based road and bridge inspection.
To address the inefficiencies, high costs, and safety risks inherent in traditional road and bridge inspection techniques, as well as the challenges posed by the large parameter volumes of current multimodal detection models and the difficulty in achieving real-time deployment on unmanned aerial vehicle (UAV) platforms, this study proposes a multimodal feature fusion road and bridge detection model based on cross distillation. The model employs a dual-branch teacher network and a single-branch student network architecture. Efficient knowledge transfer of modality-specific features is achieved through feature interaction and collaborative distillation mechanisms between the teacher networks. Concurrently, a dynamic feature fusion module, utilizing attention mechanisms, is introduced to enhance the perception of critical features associated with road and bridge defects. Experimental results demonstrate that, while maintaining a detection precision of 89.6% mAP@0.5, the proposed model reduces its parameter size to 8.2M and achieves an inference speed of 32.6 f/s. These results significantly outperform traditional multimodal fusion and lightweight methods. Compared to strategies utilizing feature concatenation or post-distillation unimodal fusion, the proposed model shows clear advantages in both detection accuracy and computational efficiency. Ablation studies confirm the effectiveness of the cross-distillation mechanism and the attention-based fusion module. The model successfully enables high-precision, lightweight detection of road and bridge defects, thus providing a technical foundation for the engineering application of UAV-based road and bridge inspection.
Available online: January 08, 2026
,DOI: 10.15888/j.cnki.csa.010105
Abstract:
Accurate identification of slope soil types is vital for stability assessment and protective design in transmission engineering. However, conventional field surveys and laboratory analyses have low efficiency and high subjectivity, thus making them difficult to satisfy the requirements for real-time identification in complex engineering scenarios. To this end, this study proposes a lightweight deep learning model based on texture integration SSR-MobileNetV2-T, which employs a dual-branch network structure combining multi-scale Gabor filtering and local binary pattern (LBP) to enhance the ability to extract soil micro-texture. Meanwhile, a multi-source soil image dataset is constructed, and the samples are enlarged via HSV-based threshold segmentation and diverse data augmentation to simulate complex field conditions and thus train the model in an end-to-end manner. The experiments show that on a five-class slope-soil image classification task, the SSR-MobileNetV2-T model achieves an average accuracy of 98.1% and an F1-score of 97.9%, generally outperforming typical lightweight models such as SVM, CNN, and EfficientNet, with prominent performance for gravel and sandy soils in particular. Parameter sensitivity analysis and ablation experiments confirm the effectiveness of each module’s design. The study indicates that SSR-MobileNetV2-T is both lightweight and highly accurate, providing efficient and reliable technical support for the intelligent identification of slope soil in transmission projects.
Accurate identification of slope soil types is vital for stability assessment and protective design in transmission engineering. However, conventional field surveys and laboratory analyses have low efficiency and high subjectivity, thus making them difficult to satisfy the requirements for real-time identification in complex engineering scenarios. To this end, this study proposes a lightweight deep learning model based on texture integration SSR-MobileNetV2-T, which employs a dual-branch network structure combining multi-scale Gabor filtering and local binary pattern (LBP) to enhance the ability to extract soil micro-texture. Meanwhile, a multi-source soil image dataset is constructed, and the samples are enlarged via HSV-based threshold segmentation and diverse data augmentation to simulate complex field conditions and thus train the model in an end-to-end manner. The experiments show that on a five-class slope-soil image classification task, the SSR-MobileNetV2-T model achieves an average accuracy of 98.1% and an F1-score of 97.9%, generally outperforming typical lightweight models such as SVM, CNN, and EfficientNet, with prominent performance for gravel and sandy soils in particular. Parameter sensitivity analysis and ablation experiments confirm the effectiveness of each module’s design. The study indicates that SSR-MobileNetV2-T is both lightweight and highly accurate, providing efficient and reliable technical support for the intelligent identification of slope soil in transmission projects.
Available online: December 31, 2025
,DOI: 10.15888/j.cnki.csa.010081
Abstract:
Cotton is an important economic crop in China, and its diseases have a significant impact on yield and quality. Therefore, it is crucial to quickly and accurately identify the types of diseases. However, existing object detection models mostly focus on improving detection accuracy while neglecting detection efficiency. These models typically have large computational requirements and a large number of parameters, making it difficult to deploy them on resource-constrained edge devices. To address this issue, this study proposes an improved YOLO11 algorithm——SDP-YOLO. The StarNet is used as the backbone network structure to reduce the number of model parameters. A DRBNCSPELAN4 module is proposed to replace C3K2 in the neck network, enhancing the semantic and positional information within the features to improve the model’s feature extraction capability. A lightweight partial convolution detection head, EPCD, is introduced to improve the model’s ability to extract important features and significantly reduce complexity. The Wise-IoU bounding box loss function is used to improve the network’s performance in bounding box regression and detection effectiveness for target diseases. Experimental results show that the improved model demonstrates significant reductions in various metrics: a 43.8% decrease in the number of parameters, a 96.9% decrease in the total floating-point operations, and a 39.6% decrease in model size, while increasing the detection accuracy by 1.3% and the FPS by 40 frames, significantly improving detection efficiency.
Cotton is an important economic crop in China, and its diseases have a significant impact on yield and quality. Therefore, it is crucial to quickly and accurately identify the types of diseases. However, existing object detection models mostly focus on improving detection accuracy while neglecting detection efficiency. These models typically have large computational requirements and a large number of parameters, making it difficult to deploy them on resource-constrained edge devices. To address this issue, this study proposes an improved YOLO11 algorithm——SDP-YOLO. The StarNet is used as the backbone network structure to reduce the number of model parameters. A DRBNCSPELAN4 module is proposed to replace C3K2 in the neck network, enhancing the semantic and positional information within the features to improve the model’s feature extraction capability. A lightweight partial convolution detection head, EPCD, is introduced to improve the model’s ability to extract important features and significantly reduce complexity. The Wise-IoU bounding box loss function is used to improve the network’s performance in bounding box regression and detection effectiveness for target diseases. Experimental results show that the improved model demonstrates significant reductions in various metrics: a 43.8% decrease in the number of parameters, a 96.9% decrease in the total floating-point operations, and a 39.6% decrease in model size, while increasing the detection accuracy by 1.3% and the FPS by 40 frames, significantly improving detection efficiency.
Available online: December 29, 2025
,DOI: 10.15888/j.cnki.csa.010088
Abstract:
Facial expression recognition has been increasingly applied in daily life. To address issues such as complex parameters, vulnerability to background interference, and high latency in facial expression recognition models, this study proposes a lightweight expression recognition method based on edge-cloud collaboration and the efficient channel attention (ECA) mechanism. A general model is deployed in the cloud and trained on large-scale datasets. Meanwhile, the shallow convolutional layers of the cloud model are transferred to the edge as feature extractors, which enhance feature extraction, generalization, and reduce the risk of overfitting. On this basis, the ECA mechanism is introduced to enable the model to focus on facial expression feature regions and suppress irrelevant information, further improving recognition accuracy and robustness. Furthermore, using depthwise separable convolution effectively reduces model parameters while maintaining expressive power, significantly lowering computational resource consumption on edge devices. Ultimately, the recognition task is performed at the edge, reducing data transmission overhead and improving response speed. Experimental results show that the accuracy of this method on the CK+ and FER2013 datasets reaches 98.76% and 71.93%, respectively. Compared with traditional methods, while maintaining high accuracy, the number of model parameters is significantly reduced and recognition latency is decreased, verifying the accuracy, efficiency, and deployment advantages of this method for facial expression recognition tasks at the edge.
Facial expression recognition has been increasingly applied in daily life. To address issues such as complex parameters, vulnerability to background interference, and high latency in facial expression recognition models, this study proposes a lightweight expression recognition method based on edge-cloud collaboration and the efficient channel attention (ECA) mechanism. A general model is deployed in the cloud and trained on large-scale datasets. Meanwhile, the shallow convolutional layers of the cloud model are transferred to the edge as feature extractors, which enhance feature extraction, generalization, and reduce the risk of overfitting. On this basis, the ECA mechanism is introduced to enable the model to focus on facial expression feature regions and suppress irrelevant information, further improving recognition accuracy and robustness. Furthermore, using depthwise separable convolution effectively reduces model parameters while maintaining expressive power, significantly lowering computational resource consumption on edge devices. Ultimately, the recognition task is performed at the edge, reducing data transmission overhead and improving response speed. Experimental results show that the accuracy of this method on the CK+ and FER2013 datasets reaches 98.76% and 71.93%, respectively. Compared with traditional methods, while maintaining high accuracy, the number of model parameters is significantly reduced and recognition latency is decreased, verifying the accuracy, efficiency, and deployment advantages of this method for facial expression recognition tasks at the edge.
Available online: December 29, 2025
,DOI: 10.15888/j.cnki.csa.010093
Abstract:
As the digital twin VR technology is increasingly widely applied, a method named RandLA-CGNet for large-scale indoor point cloud semantic segmentation is proposed to solve the problems such as the limited overall accuracy, low recognition accuracy for small objects, and blurred boundary segmentation in point cloud semantic segmentation of large-scale indoor buildings. In the encoder layer, a local-global context fusion (LGCF) module is constructed, preserving local neighborhood information while incorporating global contextual semantics. In the decoder layer, a norm-gated channel feature (NGCF) module is designed, which performs the adaptive recalibration of feature maps along the channel dimension to enhance useful information and suppress redundant noise, thereby enhancing sensitivity to details and boundaries, and improving the model’s refined recognition capability. Finally, focused cross-entropy loss (FCE loss), a hybrid loss function, is adopted to ensure stable convergence for the majority of samples and maintain overall accuracy. Additionally, this function increases the focus on hard samples and minority class samples, thereby enhancing the model’s segmentation performance in boundary regions and for rare classes. Experimental results show that the proposed model on the S3DIS dataset by employing 6-fold cross-validation increases OA, mAcc, and mIoU to 88.8%, 83.4%, and 71.9% respectively, an improvement of 0.8%, 1.4%, and 1.9% respectively compared with the baseline models. Compared to mainstream algorithms, it increases LG-Net by 0.5%, 1.0%, and 1.1% respectively, with the overall accuracy and mean intersection of union (IoU) 0.2% and 0.7% higher than FGC-AF respectively. While maintaining overall performance advantages, RandLA-CGNet improves the IoU for small objects and boundary detail segmentation by 1%–6%, significantly enhancing the recognition capability for low-frequency classes and complex boundaries. Finally, an effective solution is provided for the precise modeling of few-sample classes and detail boundaries in point cloud semantic segmentation tasks.
As the digital twin VR technology is increasingly widely applied, a method named RandLA-CGNet for large-scale indoor point cloud semantic segmentation is proposed to solve the problems such as the limited overall accuracy, low recognition accuracy for small objects, and blurred boundary segmentation in point cloud semantic segmentation of large-scale indoor buildings. In the encoder layer, a local-global context fusion (LGCF) module is constructed, preserving local neighborhood information while incorporating global contextual semantics. In the decoder layer, a norm-gated channel feature (NGCF) module is designed, which performs the adaptive recalibration of feature maps along the channel dimension to enhance useful information and suppress redundant noise, thereby enhancing sensitivity to details and boundaries, and improving the model’s refined recognition capability. Finally, focused cross-entropy loss (FCE loss), a hybrid loss function, is adopted to ensure stable convergence for the majority of samples and maintain overall accuracy. Additionally, this function increases the focus on hard samples and minority class samples, thereby enhancing the model’s segmentation performance in boundary regions and for rare classes. Experimental results show that the proposed model on the S3DIS dataset by employing 6-fold cross-validation increases OA, mAcc, and mIoU to 88.8%, 83.4%, and 71.9% respectively, an improvement of 0.8%, 1.4%, and 1.9% respectively compared with the baseline models. Compared to mainstream algorithms, it increases LG-Net by 0.5%, 1.0%, and 1.1% respectively, with the overall accuracy and mean intersection of union (IoU) 0.2% and 0.7% higher than FGC-AF respectively. While maintaining overall performance advantages, RandLA-CGNet improves the IoU for small objects and boundary detail segmentation by 1%–6%, significantly enhancing the recognition capability for low-frequency classes and complex boundaries. Finally, an effective solution is provided for the precise modeling of few-sample classes and detail boundaries in point cloud semantic segmentation tasks.
Available online: December 29, 2025
,DOI: 10.15888/j.cnki.csa.010094
Abstract:
In the noisy intermediate-scale quantum (NISQ) era, due to hardware coupling constraints, CNOT gates often cannot be executed directly. Additional SWAP gates need to be introduced to map logical qubits to suitable physical locations for ensuring circuit executability. To reduce the additional overhead caused by SWAP operations during traditional qubit mapping, this study proposes the multi-strategy quantum sparrow search algorithm (MQSSA) and applies it to qubit mapping. The qubit linkage count is defined based on the number of non-nearest neighbour (non-NN) CNOT gates acting on the same qubit pair. Combined with the physical spacing of CNOT gates, a linked quantum gate set is defined, and a fitness function is constructed based on the qubit linkage count and SWAP gate number. Simultaneously, the individual with the optimal fitness is defined as the discoverer. A quantum superposition state mechanism is introduced to equip the discoverer with parallel search capabilities, enabling the simultaneous exploration of multiple positions to expand the search space. Furthermore, to avoid falling into local optima, MQSSA introduces Gaussian noise as a follower position update perturbation mechanism, enhancing the ability to jump out of local optima. Additionally, the sentinel mechanism is implemented to maintain search diversity. Experimental results demonstrate that within the t|ket?and Qiskit compilers, MQSSA achieves average reductions of 37.5% and 46.6% in SWAP gate counts respectively, alongside average hardware overhead reductions of 13.3% and 13.2% respectively. This indicates the proposed algorithm has more efficient performance in qubit mapping, thereby improving the quality of optimization outcomes.
In the noisy intermediate-scale quantum (NISQ) era, due to hardware coupling constraints, CNOT gates often cannot be executed directly. Additional SWAP gates need to be introduced to map logical qubits to suitable physical locations for ensuring circuit executability. To reduce the additional overhead caused by SWAP operations during traditional qubit mapping, this study proposes the multi-strategy quantum sparrow search algorithm (MQSSA) and applies it to qubit mapping. The qubit linkage count is defined based on the number of non-nearest neighbour (non-NN) CNOT gates acting on the same qubit pair. Combined with the physical spacing of CNOT gates, a linked quantum gate set is defined, and a fitness function is constructed based on the qubit linkage count and SWAP gate number. Simultaneously, the individual with the optimal fitness is defined as the discoverer. A quantum superposition state mechanism is introduced to equip the discoverer with parallel search capabilities, enabling the simultaneous exploration of multiple positions to expand the search space. Furthermore, to avoid falling into local optima, MQSSA introduces Gaussian noise as a follower position update perturbation mechanism, enhancing the ability to jump out of local optima. Additionally, the sentinel mechanism is implemented to maintain search diversity. Experimental results demonstrate that within the t|ket?and Qiskit compilers, MQSSA achieves average reductions of 37.5% and 46.6% in SWAP gate counts respectively, alongside average hardware overhead reductions of 13.3% and 13.2% respectively. This indicates the proposed algorithm has more efficient performance in qubit mapping, thereby improving the quality of optimization outcomes.
Available online: December 29, 2025
,DOI: 10.15888/j.cnki.csa.010106
Abstract:
Wafer-level chips, with their enhanced integration density, superior interconnect characteristics, and lower power consumption, represent a pivotal future technology in the integrated circuit field during the post-Moore era. However, conventional simulation methods suffer from low efficiency, a lack of cross-chiplet communication modeling, and inadequate handling of heterogeneous computing resources when applied to wafer-level chips. To address the simulation requirements for wafer-level chip architectures, this study proposes a parallel discrete simulation method based on the coordination of operators and chiplets. By leveraging the coordinated parallel discrete simulation of operators and chiplets, the method effectively enhances the simulation efficiency of the system. First, a foundational standardized chiplet library and an operator library are constructed to support the architecture simulation. Subsequently, complex computation tasks are decomposed into multiple operators using the operator library, and parallel discrete simulation is realized through the collaboration of multiple cores. Communication models are incorporated to ensure the accuracy of the system simulation results. Experimental results demonstrate that compared to conventional simulation methods based on SST and Gem5, the proposed approach not only supports simulation modeling of communication between heterogeneous chiplets but also achieves an average speedup of over 4.8 times with an average accuracy loss of less than 1.3%, significantly improving the simulation efficiency for wafer-level chip systems.
Wafer-level chips, with their enhanced integration density, superior interconnect characteristics, and lower power consumption, represent a pivotal future technology in the integrated circuit field during the post-Moore era. However, conventional simulation methods suffer from low efficiency, a lack of cross-chiplet communication modeling, and inadequate handling of heterogeneous computing resources when applied to wafer-level chips. To address the simulation requirements for wafer-level chip architectures, this study proposes a parallel discrete simulation method based on the coordination of operators and chiplets. By leveraging the coordinated parallel discrete simulation of operators and chiplets, the method effectively enhances the simulation efficiency of the system. First, a foundational standardized chiplet library and an operator library are constructed to support the architecture simulation. Subsequently, complex computation tasks are decomposed into multiple operators using the operator library, and parallel discrete simulation is realized through the collaboration of multiple cores. Communication models are incorporated to ensure the accuracy of the system simulation results. Experimental results demonstrate that compared to conventional simulation methods based on SST and Gem5, the proposed approach not only supports simulation modeling of communication between heterogeneous chiplets but also achieves an average speedup of over 4.8 times with an average accuracy loss of less than 1.3%, significantly improving the simulation efficiency for wafer-level chip systems.
Available online: December 26, 2025
,DOI: 10.15888/j.cnki.csa.010082
Abstract:
To address the issues of missed detection of tiny defects, background interference, and insufficient real-time performance in surface defect detection of metal rods, an improved RT-DETR-based efficient detection algorithm, RDGS-DETR, is proposed. A lightweight feature extraction module, called the reparameterized-partial feature network block (RPFN block), is designed. It integrates structural re-parameterization with sparse channel computation to reduce parameter complexity while enhancing the representation of micro-crack features. A dynamic feature refinement fusion module (DFRFM) is also developed, incorporating the DySample dynamic upsampling operator, which improves multi-scale feature alignment accuracy in curved surface imaging scenarios by adaptively predicting offsets. Furthermore, a geometric-sensitive normalized loss (GSNL) is introduced to address the limited sensitivity of traditional IoU metrics to non-overlapping small targets and to reduce regression bias for complex-shaped defects. In addition, a sparse global interaction attention (SGIA) module is developed, which employs an efficient additive attention mechanism to achieve global context modeling of defect regions with linear complexity. Experimental results demonstrate that, compared with the original model, RDGS-DETR improves inference speed by 8.55 FPS and increases mAP@0.5 by 2.8%, while also verifying its robustness. The algorithm achieves a balance between detection accuracy and real-time performance, providing reliable technical support for surface quality inspection of metal rods in intelligent manufacturing scenarios.
To address the issues of missed detection of tiny defects, background interference, and insufficient real-time performance in surface defect detection of metal rods, an improved RT-DETR-based efficient detection algorithm, RDGS-DETR, is proposed. A lightweight feature extraction module, called the reparameterized-partial feature network block (RPFN block), is designed. It integrates structural re-parameterization with sparse channel computation to reduce parameter complexity while enhancing the representation of micro-crack features. A dynamic feature refinement fusion module (DFRFM) is also developed, incorporating the DySample dynamic upsampling operator, which improves multi-scale feature alignment accuracy in curved surface imaging scenarios by adaptively predicting offsets. Furthermore, a geometric-sensitive normalized loss (GSNL) is introduced to address the limited sensitivity of traditional IoU metrics to non-overlapping small targets and to reduce regression bias for complex-shaped defects. In addition, a sparse global interaction attention (SGIA) module is developed, which employs an efficient additive attention mechanism to achieve global context modeling of defect regions with linear complexity. Experimental results demonstrate that, compared with the original model, RDGS-DETR improves inference speed by 8.55 FPS and increases mAP@0.5 by 2.8%, while also verifying its robustness. The algorithm achieves a balance between detection accuracy and real-time performance, providing reliable technical support for surface quality inspection of metal rods in intelligent manufacturing scenarios.
Available online: December 26, 2025
,DOI: 10.15888/j.cnki.csa.010083
Abstract:
In recent years, federated learning (FL) has emerged as a distributed machine learning paradigm that enables model training while preserving data privacy. It has been widely applied in domains such as smart healthcare, financial services, the Internet of Things (IoT), and the Internet of Vehicles (IoV). However, due to the highly dynamic nature of IoV environments and the heterogeneous computing resources among vehicles, not all clients are suitable for participation in federated training. Therefore, designing an efficient and robust client selection strategy is critical for ensuring model performance and system efficiency. Traditional FL methods often rely on static or heuristic client selection mechanisms, which fail to adapt to the frequently changing states and characteristics of clients in IoV scenarios. To address this issue, this study proposes a dynamic client selection approach based on entropy regularized proximal policy optimization (ERPPO), integrated with a confidence-weighted aggregation mechanism. By incorporating a policy entropy regularization term into the PPO objective function, the proposed method enhances the exploration capability of the client selection policy, thus mitigating the risk of local optima. Furthermore, the confidence-based aggregation strategy adaptively adjusts the aggregation weights based on the variance of local model updates, which enhances the convergence stability and robustness of the global model. Experimental results demonstrate that the proposed ERPPO framework not only reduces communication overhead but also achieves superior overall performance in dynamic environments while maintaining high model precision.
In recent years, federated learning (FL) has emerged as a distributed machine learning paradigm that enables model training while preserving data privacy. It has been widely applied in domains such as smart healthcare, financial services, the Internet of Things (IoT), and the Internet of Vehicles (IoV). However, due to the highly dynamic nature of IoV environments and the heterogeneous computing resources among vehicles, not all clients are suitable for participation in federated training. Therefore, designing an efficient and robust client selection strategy is critical for ensuring model performance and system efficiency. Traditional FL methods often rely on static or heuristic client selection mechanisms, which fail to adapt to the frequently changing states and characteristics of clients in IoV scenarios. To address this issue, this study proposes a dynamic client selection approach based on entropy regularized proximal policy optimization (ERPPO), integrated with a confidence-weighted aggregation mechanism. By incorporating a policy entropy regularization term into the PPO objective function, the proposed method enhances the exploration capability of the client selection policy, thus mitigating the risk of local optima. Furthermore, the confidence-based aggregation strategy adaptively adjusts the aggregation weights based on the variance of local model updates, which enhances the convergence stability and robustness of the global model. Experimental results demonstrate that the proposed ERPPO framework not only reduces communication overhead but also achieves superior overall performance in dynamic environments while maintaining high model precision.
Available online: December 26, 2025
,DOI: 10.15888/j.cnki.csa.010070
Abstract:
To address the challenges posed by dynamic environments to simultaneous localization and mapping (SLAM), this study proposes a detection-first tightly-coupled LiDAR-visual-inertial SLAM system that integrates a LiDAR, a camera, and an inertial measurement unit (IMU). First, semantically labeled point-cloud clusters are obtained through the fusion of images and point-cloud information. Then, a tracking algorithm is used to acquire the motion state information of targets. Subsequently, the tracked dynamic targets are utilized to eliminate redundant feature points. Finally, a factor graph is adopted to jointly optimize IMU pre-integration and achieve tight coupling between LiDAR and visual odometry. To validate the performance of the proposed SLAM framework, experiments are conducted on both public datasets (KITTI and UrbanNav) and real-world data. Experimental results demonstrate that in highly dynamic and normal scenarios in public datasets, compared with the LeGO-LOAM, LIO-SAM, and LVI-SAM algorithms, the proposed algorithm reduces the root mean square error (RMSE) by 44.56% (4.47 m) and 4.15% (4.62 m), respectively. Real-world testing confirms that the algorithm effectively mitigates the direct impact of dynamic objects on map construction.
To address the challenges posed by dynamic environments to simultaneous localization and mapping (SLAM), this study proposes a detection-first tightly-coupled LiDAR-visual-inertial SLAM system that integrates a LiDAR, a camera, and an inertial measurement unit (IMU). First, semantically labeled point-cloud clusters are obtained through the fusion of images and point-cloud information. Then, a tracking algorithm is used to acquire the motion state information of targets. Subsequently, the tracked dynamic targets are utilized to eliminate redundant feature points. Finally, a factor graph is adopted to jointly optimize IMU pre-integration and achieve tight coupling between LiDAR and visual odometry. To validate the performance of the proposed SLAM framework, experiments are conducted on both public datasets (KITTI and UrbanNav) and real-world data. Experimental results demonstrate that in highly dynamic and normal scenarios in public datasets, compared with the LeGO-LOAM, LIO-SAM, and LVI-SAM algorithms, the proposed algorithm reduces the root mean square error (RMSE) by 44.56% (4.47 m) and 4.15% (4.62 m), respectively. Real-world testing confirms that the algorithm effectively mitigates the direct impact of dynamic objects on map construction.
Available online: December 26, 2025
,DOI: 10.15888/j.cnki.csa.010075
Abstract:
In response to the challenges of resource constraints and high load in roadside unit (RSU) within vehicular edge computing (VEC), as well as the limitations of existing task offloading optimization schemes that focus solely on reducing latency or energy consumption while neglecting the security issues faced by edge nodes, this study proposes a task offloading scheme based on trust awareness and the proximal policy optimization (PPO) algorithm. First, a VEC network architecture is constructed, which utilizes the computing resources of nearby idle vehicles to process tasks locally or offload them to RSU or idle service vehicles, in order to reduce the overall system latency and energy consumption. Second, a dynamic feedback trust evaluation model based on multi-source weighting and a reward-punishment mechanism is constructed to achieve a quantitative assessment of the trustworthiness of edge nodes. Finally, the task offloading strategy is optimized using the PPO algorithm based on deep reinforcement learning. The experimental results show that compared to the DQN, D3QN, and TASACO algorithms, the proposed scheme has better convergence and stability, and it outperforms existing schemes in terms of task execution latency and energy consumption.
In response to the challenges of resource constraints and high load in roadside unit (RSU) within vehicular edge computing (VEC), as well as the limitations of existing task offloading optimization schemes that focus solely on reducing latency or energy consumption while neglecting the security issues faced by edge nodes, this study proposes a task offloading scheme based on trust awareness and the proximal policy optimization (PPO) algorithm. First, a VEC network architecture is constructed, which utilizes the computing resources of nearby idle vehicles to process tasks locally or offload them to RSU or idle service vehicles, in order to reduce the overall system latency and energy consumption. Second, a dynamic feedback trust evaluation model based on multi-source weighting and a reward-punishment mechanism is constructed to achieve a quantitative assessment of the trustworthiness of edge nodes. Finally, the task offloading strategy is optimized using the PPO algorithm based on deep reinforcement learning. The experimental results show that compared to the DQN, D3QN, and TASACO algorithms, the proposed scheme has better convergence and stability, and it outperforms existing schemes in terms of task execution latency and energy consumption.
Available online: December 26, 2025
,DOI: 10.15888/j.cnki.csa.010076
Abstract:
With the continuous advancement of smart city development, safety issues in building edge areas have become increasingly severe, as incidents of accidental falls and falling objects occur frequently. There is an urgent need for more intelligent and efficient monitoring solutions. To address the limited temporal modeling capabilities of current object detection methods, particularly in recognizing small, occluded, and fast-moving targets, this study proposes a video detection framework that integrates multiple temporal semantic enhancement mechanisms for the unified detection of both people and falling objects. The proposed method is built upon a faster R-CNN backbone and incorporates three temporal-aware modules: motion-aware module (MAM), temporal region of interest align (TROI Align), and sequence-level semantic aggregation head (SELSA Head). These modules enhance the model’s perception of dynamic objects in complex temporal scenarios from three perspectives: motion saliency modeling, spatial alignment, and semantic aggregation. To support model training and evaluation, a dedicated video dataset covering multiple building edge scenarios and various types of risk targets is constructed. Experimental results demonstrate that the proposed method achieves strong performance in both “detection of personnel behavior at building edges” and “falling object detection” tasks, showing excellent cross-task robustness and practical application potential.
With the continuous advancement of smart city development, safety issues in building edge areas have become increasingly severe, as incidents of accidental falls and falling objects occur frequently. There is an urgent need for more intelligent and efficient monitoring solutions. To address the limited temporal modeling capabilities of current object detection methods, particularly in recognizing small, occluded, and fast-moving targets, this study proposes a video detection framework that integrates multiple temporal semantic enhancement mechanisms for the unified detection of both people and falling objects. The proposed method is built upon a faster R-CNN backbone and incorporates three temporal-aware modules: motion-aware module (MAM), temporal region of interest align (TROI Align), and sequence-level semantic aggregation head (SELSA Head). These modules enhance the model’s perception of dynamic objects in complex temporal scenarios from three perspectives: motion saliency modeling, spatial alignment, and semantic aggregation. To support model training and evaluation, a dedicated video dataset covering multiple building edge scenarios and various types of risk targets is constructed. Experimental results demonstrate that the proposed method achieves strong performance in both “detection of personnel behavior at building edges” and “falling object detection” tasks, showing excellent cross-task robustness and practical application potential.
Available online: December 26, 2025
,DOI: 10.15888/j.cnki.csa.010078
Abstract:
Diverse degradation types and the difficulty of detail recovery make real-world image super-resolution challenging, with existing methods still struggling with structural preservation and semantic consistency. This study proposes a semantic-aware interactive diffusion method for image super-resolution reconstruction (SISRM) method. Semantic segmentation information is introduced as prior knowledge, enhancing structural understanding and providing semantic guidance during reconstruction. Specifically, a segmentation-aware prompt extractor is designed and trained to efficiently obtain segmentation mask embeddings and semantic labels from degraded low-resolution images using a segmentation mask encoder and a label text generator. An interactive text-to-image controller is then introduced, integrating a segmentation-guided cross-attention module with a trainable image encoder. The diffusion process is guided under multi-modal semantic conditions to enhance local detail and global structure awareness. Finally, a mask feature fusion mechanism is proposed to mitigate the mismatch between local conditional control and the global latent distribution, improving the consistency and visual quality of the generated images. Experimental results on the DIV2K-Val and RealSR datasets show that the proposed method achieves the highest scores of 0.6121 and 0.7274 in no-reference and cross-modal image quality assessment, respectively. These results demonstrate notable improvements in detail restoration, semantic consistency, and overall perceptual quality.
Diverse degradation types and the difficulty of detail recovery make real-world image super-resolution challenging, with existing methods still struggling with structural preservation and semantic consistency. This study proposes a semantic-aware interactive diffusion method for image super-resolution reconstruction (SISRM) method. Semantic segmentation information is introduced as prior knowledge, enhancing structural understanding and providing semantic guidance during reconstruction. Specifically, a segmentation-aware prompt extractor is designed and trained to efficiently obtain segmentation mask embeddings and semantic labels from degraded low-resolution images using a segmentation mask encoder and a label text generator. An interactive text-to-image controller is then introduced, integrating a segmentation-guided cross-attention module with a trainable image encoder. The diffusion process is guided under multi-modal semantic conditions to enhance local detail and global structure awareness. Finally, a mask feature fusion mechanism is proposed to mitigate the mismatch between local conditional control and the global latent distribution, improving the consistency and visual quality of the generated images. Experimental results on the DIV2K-Val and RealSR datasets show that the proposed method achieves the highest scores of 0.6121 and 0.7274 in no-reference and cross-modal image quality assessment, respectively. These results demonstrate notable improvements in detail restoration, semantic consistency, and overall perceptual quality.
Available online: December 19, 2025
,DOI: 10.15888/j.cnki.csa.010072
Abstract:
In the field of infrared small target detection for unmanned aerial vehicle (UAV), complex ground backgrounds and the inherently small size of targets often result in missed detections or false alarms in detection models. To address this problem, this study proposes a lightweight and high-precision infrared small target detection algorithm based on the YOLOv7 framework, termed unmanned aerial vehicle-you only look once (UAV-YOLO). First, given that the targets of interest are predominantly small, the 1×1 convolutions in the ELAN, ELAN-W, and CARAFE modules of the YOLOv7 backbone network, as well as those in the neck network, are replaced with GSConv. At the same time, the P5 detection head, which provides limited efficiency in infrared small target detection, is removed, and a newly added P2 detection head is designed specifically for small targets. These modifications not only enhance detection efficiency but also substantially reduce the number of parameters, thus achieving a lightweight model. Second, an improved SPPFCSPC pyramid pooling module is integrated into the backbone network. The inclusion of this module effectively expands the receptive field of the model, thus improving the detection accuracy of infrared small targets. Subsequently, the content-aware reassembly of feature (CARAFE) module is incorporated into YOLOv7. This module enhances the preservation and optimization of feature representations for small targets. Furthermore, a coordinate attention (CA) mechanism is introduced before the detection head, enabling more precise localization of small targets and allowing the detection head to focus on key regions. Finally, the normalized Wasserstein distance (NWD) loss is adopted to replace the CIoU loss. This replacement reduces the sensitivity of the model to positional deviations and further improves detection performance. Experimental results demonstrate that compared with the baseline YOLOv7 model, the proposed UAV-YOLO achieves an mAP of 95.7%, representing a 5.2% improvement, while the number of parameters is reduced to 12.0M, a decrease of 67.7%. These improvements ensure high accuracy while significantly reducing the number of parameters, thus verifying the effectiveness and practicality of the proposed infrared small target detection model for UAV applications.
In the field of infrared small target detection for unmanned aerial vehicle (UAV), complex ground backgrounds and the inherently small size of targets often result in missed detections or false alarms in detection models. To address this problem, this study proposes a lightweight and high-precision infrared small target detection algorithm based on the YOLOv7 framework, termed unmanned aerial vehicle-you only look once (UAV-YOLO). First, given that the targets of interest are predominantly small, the 1×1 convolutions in the ELAN, ELAN-W, and CARAFE modules of the YOLOv7 backbone network, as well as those in the neck network, are replaced with GSConv. At the same time, the P5 detection head, which provides limited efficiency in infrared small target detection, is removed, and a newly added P2 detection head is designed specifically for small targets. These modifications not only enhance detection efficiency but also substantially reduce the number of parameters, thus achieving a lightweight model. Second, an improved SPPFCSPC pyramid pooling module is integrated into the backbone network. The inclusion of this module effectively expands the receptive field of the model, thus improving the detection accuracy of infrared small targets. Subsequently, the content-aware reassembly of feature (CARAFE) module is incorporated into YOLOv7. This module enhances the preservation and optimization of feature representations for small targets. Furthermore, a coordinate attention (CA) mechanism is introduced before the detection head, enabling more precise localization of small targets and allowing the detection head to focus on key regions. Finally, the normalized Wasserstein distance (NWD) loss is adopted to replace the CIoU loss. This replacement reduces the sensitivity of the model to positional deviations and further improves detection performance. Experimental results demonstrate that compared with the baseline YOLOv7 model, the proposed UAV-YOLO achieves an mAP of 95.7%, representing a 5.2% improvement, while the number of parameters is reduced to 12.0M, a decrease of 67.7%. These improvements ensure high accuracy while significantly reducing the number of parameters, thus verifying the effectiveness and practicality of the proposed infrared small target detection model for UAV applications.
Available online: December 19, 2025
,DOI: 10.15888/j.cnki.csa.010065
Abstract:
As an innovative renewable and clean energy device, the proton exchange membrane fuel cell (PEMFC) holds immense market application value. PEMFCs are susceptible to water management faults during prolonged operation under complex and varying conditions. However, traditional fault diagnosis methods struggle to effectively extract key fault features from dynamically changing monitoring data. To address this, this study proposes a PEMFC fault diagnosis method based on a deep parallel residual neural network (DP-ResNet). This method initially processes the collected multi-source signals, such as current and voltage. Subsequently, a DP-ResNet is designed to overcome the limitation of residual networks in multi-scale feature extraction. Finally, the proposed algorithm is applied to a dataset of PEMFC water management faults under varying load conditions for diagnostic verification. Experimental results demonstrate that the proposed DP-ResNet model achieves a diagnostic accuracy of up to 99.46% for flooding faults in real PEMFC experimental datasets. Compared with traditional machine learning algorithms such as Decision-tree, GaussianNB, KNN, and CNN, the proposed method demonstrates superior feature extraction and diagnostic accuracy.
As an innovative renewable and clean energy device, the proton exchange membrane fuel cell (PEMFC) holds immense market application value. PEMFCs are susceptible to water management faults during prolonged operation under complex and varying conditions. However, traditional fault diagnosis methods struggle to effectively extract key fault features from dynamically changing monitoring data. To address this, this study proposes a PEMFC fault diagnosis method based on a deep parallel residual neural network (DP-ResNet). This method initially processes the collected multi-source signals, such as current and voltage. Subsequently, a DP-ResNet is designed to overcome the limitation of residual networks in multi-scale feature extraction. Finally, the proposed algorithm is applied to a dataset of PEMFC water management faults under varying load conditions for diagnostic verification. Experimental results demonstrate that the proposed DP-ResNet model achieves a diagnostic accuracy of up to 99.46% for flooding faults in real PEMFC experimental datasets. Compared with traditional machine learning algorithms such as Decision-tree, GaussianNB, KNN, and CNN, the proposed method demonstrates superior feature extraction and diagnostic accuracy.
Available online: December 19, 2025
,DOI: 10.15888/j.cnki.csa.010066
Abstract:
The widespread adoption of encryption technology has given malicious activities a chance to hide, posing a great challenge to network security monitoring systems. Existing encrypted traffic detection methods primarily extract statistical traffic features at the individual packet level. However, this may disrupt the features implied in the original continuous communication behavior, due to potential IP fragmentation. Furthermore, most approaches model the interaction patterns between network flows at a relatively coarse granularity, failing to thoroughly explore the communication intent between peer entities. This study introduces a novel method, interaction state graph-net (ISG-Net), which uses interaction as the analysis granularity. ISG-Net constructs a traffic interaction state graph based on state transitions and applies a self-attentive encoder model to capture temporal traffic information. In particular, interaction state representations containing global information are obtained through the interaction state graph. Then, fine-grained features of each interaction are extracted to obtain the representation of the sessions (bidirectional flows). Experiments on three datasets demonstrate that the proposed method outperforms existing methods in terms of accuracy, robustness and fault tolerance in the task of encrypted malicious traffic detection.
The widespread adoption of encryption technology has given malicious activities a chance to hide, posing a great challenge to network security monitoring systems. Existing encrypted traffic detection methods primarily extract statistical traffic features at the individual packet level. However, this may disrupt the features implied in the original continuous communication behavior, due to potential IP fragmentation. Furthermore, most approaches model the interaction patterns between network flows at a relatively coarse granularity, failing to thoroughly explore the communication intent between peer entities. This study introduces a novel method, interaction state graph-net (ISG-Net), which uses interaction as the analysis granularity. ISG-Net constructs a traffic interaction state graph based on state transitions and applies a self-attentive encoder model to capture temporal traffic information. In particular, interaction state representations containing global information are obtained through the interaction state graph. Then, fine-grained features of each interaction are extracted to obtain the representation of the sessions (bidirectional flows). Experiments on three datasets demonstrate that the proposed method outperforms existing methods in terms of accuracy, robustness and fault tolerance in the task of encrypted malicious traffic detection.
Available online: December 19, 2025
,DOI: 10.15888/j.cnki.csa.010058
Abstract:
With the rapid development of large language models (LLMs), their application in the explainability of recommender systems has become a research hotspot. This study systematically reviews the research progress of LLMs in the explainability of recommender systems, providing a comprehensive overview covering current research status, evaluation metrics, datasets, and application scenarios. From a technical perspective, existing research is categorized into LLM-based recommender systems and LLM-aid recommender systems, further subdivided according to whether fine-tuning is required. In terms of evaluation metrics, manual evaluation and automated evaluation metrics are summarized, with automated evaluation metrics including traditional metrics, LLM-integrated metrics, and extended metrics. Moreover, the usage of public and private datasets is reviewed, with emphasis on the importance of review data in explainable recommendations. Finally, practical applications of LLMs in the explainability of recommender systems across various domains are explored, and the challenges faced by current research as well as potential future research directions are analyzed.
With the rapid development of large language models (LLMs), their application in the explainability of recommender systems has become a research hotspot. This study systematically reviews the research progress of LLMs in the explainability of recommender systems, providing a comprehensive overview covering current research status, evaluation metrics, datasets, and application scenarios. From a technical perspective, existing research is categorized into LLM-based recommender systems and LLM-aid recommender systems, further subdivided according to whether fine-tuning is required. In terms of evaluation metrics, manual evaluation and automated evaluation metrics are summarized, with automated evaluation metrics including traditional metrics, LLM-integrated metrics, and extended metrics. Moreover, the usage of public and private datasets is reviewed, with emphasis on the importance of review data in explainable recommendations. Finally, practical applications of LLMs in the explainability of recommender systems across various domains are explored, and the challenges faced by current research as well as potential future research directions are analyzed.
Available online: December 19, 2025
,DOI: 10.15888/j.cnki.csa.010059
Abstract:
In view of the problem that the existing reconstruction methods have no ideal effect due to the texture complexity during the super-resolution reconstruction of rock thin slice images, this study proposes a super-resolution denoising diffusion probability model of rock slice (rsDDPMSR). To solve the problem that the traditional upsampling methods tend to cause artifacts and insufficient utilization of prior information of low-resolution images, it puts forward a layered feature enhancement network (LFE-Net). Meanwhile, a dual-path network is employed to conduct layered feature enhancement on the high-frequency and low-frequency components decomposed by the stationary wavelet transform. The low-resolution features enhanced by LFE-Net are combined with the feature channels of the target high-resolution noisy image as the conditional input of the diffusion model to guide the generation direction of the diffusion model and provide rich prior information. Based on U-Net, a double-encoder multi-scale noise prediction network (ACA-U-Net) is designed to effectively process multi-scale information of rock slices, and a time-aware adaptive cross-attention mechanism is introduced into the skip connections to match the feature distribution changes in different denoising stages of the diffusion model to enhance the model’s attention to key areas and effectively improve image reconstruction details. The experimental results show that rsDDPMSR has a better reconstruction effect at 2×, 4×, and 8× magnification than other mainstream reconstruction methods, such as CAMixerSR, SDFlow, IDM, and SR3.
In view of the problem that the existing reconstruction methods have no ideal effect due to the texture complexity during the super-resolution reconstruction of rock thin slice images, this study proposes a super-resolution denoising diffusion probability model of rock slice (rsDDPMSR). To solve the problem that the traditional upsampling methods tend to cause artifacts and insufficient utilization of prior information of low-resolution images, it puts forward a layered feature enhancement network (LFE-Net). Meanwhile, a dual-path network is employed to conduct layered feature enhancement on the high-frequency and low-frequency components decomposed by the stationary wavelet transform. The low-resolution features enhanced by LFE-Net are combined with the feature channels of the target high-resolution noisy image as the conditional input of the diffusion model to guide the generation direction of the diffusion model and provide rich prior information. Based on U-Net, a double-encoder multi-scale noise prediction network (ACA-U-Net) is designed to effectively process multi-scale information of rock slices, and a time-aware adaptive cross-attention mechanism is introduced into the skip connections to match the feature distribution changes in different denoising stages of the diffusion model to enhance the model’s attention to key areas and effectively improve image reconstruction details. The experimental results show that rsDDPMSR has a better reconstruction effect at 2×, 4×, and 8× magnification than other mainstream reconstruction methods, such as CAMixerSR, SDFlow, IDM, and SR3.
Available online: December 19, 2025
,DOI: 10.15888/j.cnki.csa.010092
Abstract:
Compared to traditional 3D reconstruction methods, neural radiance fields (NeRFs) can effectively capture implicit neural representations, enabling high-quality 3D reconstruction and novel view synthesis tasks. However, NeRFs typically require a large amount of raw data for training. To this end, this study proposes a method by integrating variational autoencoders (VAEs) with NeRF to improve 3D scene generation performance under limited training data. Firstly, by constructing a VAE encoder, a certain proportion of raw images from the training data are selected to form a vector set. Meanwhile, the encoder compresses this set to capture latent feature vectors, which are then employed to supplement global scene information in the input layer. Secondly, an adaptive-enhanced sampling algorithm is developed to dynamically adjust the distribution density of sampling points, thereby improving NeRF’s ability to capture fine scene details. Experiments conducted on three public datasets demonstrate the effectiveness of the proposed method. Additionally, the proposed method achieves 3D reconstruction results comparable to baseline methods trained with full datasets, even under the loss of original training data.
Compared to traditional 3D reconstruction methods, neural radiance fields (NeRFs) can effectively capture implicit neural representations, enabling high-quality 3D reconstruction and novel view synthesis tasks. However, NeRFs typically require a large amount of raw data for training. To this end, this study proposes a method by integrating variational autoencoders (VAEs) with NeRF to improve 3D scene generation performance under limited training data. Firstly, by constructing a VAE encoder, a certain proportion of raw images from the training data are selected to form a vector set. Meanwhile, the encoder compresses this set to capture latent feature vectors, which are then employed to supplement global scene information in the input layer. Secondly, an adaptive-enhanced sampling algorithm is developed to dynamically adjust the distribution density of sampling points, thereby improving NeRF’s ability to capture fine scene details. Experiments conducted on three public datasets demonstrate the effectiveness of the proposed method. Additionally, the proposed method achieves 3D reconstruction results comparable to baseline methods trained with full datasets, even under the loss of original training data.
Available online: November 26, 2025
,DOI: 10.15888/j.cnki.csa.010063
Abstract:
Large language models (LLMs), represented by ChatGPT and DeepSeek, are rapidly developing and widely used in various tasks, such as text generation and intelligent assistants. However, these large models also face severe privacy and security risks. Especially in high security scenarios such as healthcare and finance, threats such as model theft and data privacy leakage are often key factors hindering the application of large models. Existing security solutions for protecting large model inference usually have certain limitations, such as the lack of runtime protection for the inference computation process, or practical challenges caused by the high cost of computation and communication. Confidential computing can build a secure inference environment based on trusted execution environment (TEE) hardware, and is a practical and effective security technology for implementing secure inference of large language models. Therefore, this study proposes a secure inference application scheme for large language models based on confidential computing, which ensures the integrity of the inference computing environment, model weight parameters, and model image files through remote attestation, implements encryption protection for large model inference traffic via confidential interconnection based on TEE hardware, and protects the privacy of user prompts in multi-user scenarios by isolating the inference contexts among different users. The proposed scheme provides comprehensive security protection for the entire process and full chain of large language model inference, while verifying the integrity of the execution environment to achieve efficient and secure confidential large language model inference. Furthermore, a prototype system is implemented on a heterogeneous TEE server platform (SEV and CSV), and the system’s security and performance are evaluated. The results show that while achieving the expected security goals, the performance loss introduced by the proposed scheme theoretically does not exceed 1% of the inference overhead of the native AI model, which can be ignored in practical applications.
Large language models (LLMs), represented by ChatGPT and DeepSeek, are rapidly developing and widely used in various tasks, such as text generation and intelligent assistants. However, these large models also face severe privacy and security risks. Especially in high security scenarios such as healthcare and finance, threats such as model theft and data privacy leakage are often key factors hindering the application of large models. Existing security solutions for protecting large model inference usually have certain limitations, such as the lack of runtime protection for the inference computation process, or practical challenges caused by the high cost of computation and communication. Confidential computing can build a secure inference environment based on trusted execution environment (TEE) hardware, and is a practical and effective security technology for implementing secure inference of large language models. Therefore, this study proposes a secure inference application scheme for large language models based on confidential computing, which ensures the integrity of the inference computing environment, model weight parameters, and model image files through remote attestation, implements encryption protection for large model inference traffic via confidential interconnection based on TEE hardware, and protects the privacy of user prompts in multi-user scenarios by isolating the inference contexts among different users. The proposed scheme provides comprehensive security protection for the entire process and full chain of large language model inference, while verifying the integrity of the execution environment to achieve efficient and secure confidential large language model inference. Furthermore, a prototype system is implemented on a heterogeneous TEE server platform (SEV and CSV), and the system’s security and performance are evaluated. The results show that while achieving the expected security goals, the performance loss introduced by the proposed scheme theoretically does not exceed 1% of the inference overhead of the native AI model, which can be ignored in practical applications.
Available online: November 26, 2025
,DOI: 10.15888/j.cnki.csa.010071
Abstract:
Precipitation nowcasting, a critical spatiotemporal sequence prediction task, has significant applications in meteorological domains such as agriculture and transportation. While radar echo extrapolation based on deep learning is a commonly used nowcasting method, existing methods have limitations in capturing the complex spatiotemporal patterns of radar echoes. The performance of these methods degrades significantly over time, making it difficult to accurately predict the spatiotemporal evolution of precipitation. This study proposes GloCal-Net, a model that integrates global modes and local variations. The model is based on a U-Net architecture with hybrid Mamba-Transformer experts, designed to enhance the ability to capture complex patterns in radar echo sequences by optimizing the feature extraction mechanism. To validate the proposed model, comparative and ablation experiments are conducted on a real radar dataset from Jiujiang. Compared with mainstream deep learning models, in the 2-hour extrapolation task, the proposed model achieves a comparable Heidke skill score and a 4.19% higher critical success index, reaching 0.36 and 0.29 respectively. The learned perceptual image patch similarity decreases by 3.70%, reaching 0.31. The structural similarity increases by 2.07%, reaching 72.37%. These experimental results show that GloCal-Net improves several key performance indicators and simultaneously verifies the effectiveness of each component.
Precipitation nowcasting, a critical spatiotemporal sequence prediction task, has significant applications in meteorological domains such as agriculture and transportation. While radar echo extrapolation based on deep learning is a commonly used nowcasting method, existing methods have limitations in capturing the complex spatiotemporal patterns of radar echoes. The performance of these methods degrades significantly over time, making it difficult to accurately predict the spatiotemporal evolution of precipitation. This study proposes GloCal-Net, a model that integrates global modes and local variations. The model is based on a U-Net architecture with hybrid Mamba-Transformer experts, designed to enhance the ability to capture complex patterns in radar echo sequences by optimizing the feature extraction mechanism. To validate the proposed model, comparative and ablation experiments are conducted on a real radar dataset from Jiujiang. Compared with mainstream deep learning models, in the 2-hour extrapolation task, the proposed model achieves a comparable Heidke skill score and a 4.19% higher critical success index, reaching 0.36 and 0.29 respectively. The learned perceptual image patch similarity decreases by 3.70%, reaching 0.31. The structural similarity increases by 2.07%, reaching 72.37%. These experimental results show that GloCal-Net improves several key performance indicators and simultaneously verifies the effectiveness of each component.
Available online: November 26, 2025
,DOI: 10.15888/j.cnki.csa.010077
Abstract:
The agile requirements process model is suitable for scenarios with frequent requirements iterations. This approach emphasizes a user-centered design concept, using concise text and not relying on complex processes and tools. Introducing requirement models into the agile process can effectively address issues such as the insufficient understanding of agile methods. However, in scenarios with frequent requirement iterations, the introduced requirement models often face challenges such as difficulty in maintenance and outdated versions. In agile development with frequent requirements iterations, the model’s complexity results in high resource consumption for its manual maintenance of the requirement model. To address this issue, this study proposes an agile requirements process model based on multi-agent systems, MA-ARP. This model uses an automatic processing system built around multi-agent technology, leveraging the reasoning and recognition capabilities of multi-agents to dynamically update the requirement model according to changes in requirements. This approach effectively reduces the costs associated with maintaining the requirement model during the agile process. Case studies and comprehensive evaluations show that this approach can achieve automatic updates and maintenance of the requirement model, and the proposed model meets or exceeds level 2 in most of the selected requirements engineering process evaluation metrics.
The agile requirements process model is suitable for scenarios with frequent requirements iterations. This approach emphasizes a user-centered design concept, using concise text and not relying on complex processes and tools. Introducing requirement models into the agile process can effectively address issues such as the insufficient understanding of agile methods. However, in scenarios with frequent requirement iterations, the introduced requirement models often face challenges such as difficulty in maintenance and outdated versions. In agile development with frequent requirements iterations, the model’s complexity results in high resource consumption for its manual maintenance of the requirement model. To address this issue, this study proposes an agile requirements process model based on multi-agent systems, MA-ARP. This model uses an automatic processing system built around multi-agent technology, leveraging the reasoning and recognition capabilities of multi-agents to dynamically update the requirement model according to changes in requirements. This approach effectively reduces the costs associated with maintaining the requirement model during the agile process. Case studies and comprehensive evaluations show that this approach can achieve automatic updates and maintenance of the requirement model, and the proposed model meets or exceeds level 2 in most of the selected requirements engineering process evaluation metrics.
Available online: November 17, 2025
,DOI: 10.15888/j.cnki.csa.010041
Abstract:
Multi-agent reinforcement learning (MARL) is a crucial part of multi-agent system research, demonstrating remarkable effectiveness in complex collaborative tasks. However, in scenarios requiring long-term decision-making, multi-agent systems often underperform due to the difficulty in estimating long-term returns and accurately modeling environmental uncertainties. To this end, this study proposes a multi-agent memory-reinforcement learning model based on quantile regression. The model not only selectively utilizes historical decision-making experience to assist long-term decision-making but also employs quantile functions to model the return distribution, thereby effectively capturing return uncertainties. The model comprises three components, including a memory indexing module, an implicit quantile decision network, and a value distribution decomposition module. Specifically, the memory indexing module generates intrinsic reward by adopting historical decision-making experience to enhance the agents’ full utilization of existing experience. The implicit quantile decision network models reward distribution via quantile regression, providing powerful support for long-term decision-making. The value distribution decomposition module decomposes overall return distributions into the distribution of an individual agent to support single-agent strategy learning. Extensive experiments conducted in StarCraft II environments demonstrate that the proposed method enhances the performance of agents in long-term decision-making tasks, with fast convergence rates.
Multi-agent reinforcement learning (MARL) is a crucial part of multi-agent system research, demonstrating remarkable effectiveness in complex collaborative tasks. However, in scenarios requiring long-term decision-making, multi-agent systems often underperform due to the difficulty in estimating long-term returns and accurately modeling environmental uncertainties. To this end, this study proposes a multi-agent memory-reinforcement learning model based on quantile regression. The model not only selectively utilizes historical decision-making experience to assist long-term decision-making but also employs quantile functions to model the return distribution, thereby effectively capturing return uncertainties. The model comprises three components, including a memory indexing module, an implicit quantile decision network, and a value distribution decomposition module. Specifically, the memory indexing module generates intrinsic reward by adopting historical decision-making experience to enhance the agents’ full utilization of existing experience. The implicit quantile decision network models reward distribution via quantile regression, providing powerful support for long-term decision-making. The value distribution decomposition module decomposes overall return distributions into the distribution of an individual agent to support single-agent strategy learning. Extensive experiments conducted in StarCraft II environments demonstrate that the proposed method enhances the performance of agents in long-term decision-making tasks, with fast convergence rates.
Available online: November 17, 2025
,DOI: 10.15888/j.cnki.csa.010060
Abstract:
As an important therapeutic resource, traditional Chinese medicine (TCM) has undergone thousands of years of clinical practice and application. To promote the modernization of TCM and explore its application potential in new indications, this study draws on research experience from drug repurposing in Western medicine and combines emerging network medicine theories to propose two random walk-based models for predicting potential therapeutic associations between TCM and symptoms: M-RW and GO-DREAMwalk. The two models incorporate path-based and functional information between TCM and symptoms to guide the random walk process. The resulting node sequences are input into a heterogeneous Skip-gram model to learn the embedded vector representations of nodes. Subsequently, an XGBoost classifier is trained by adopting TCM-symptom association labels and the learned embedded vectors. Finally, the models are tested and evaluated by employing clinical data on liver cirrhosis. In the clinically effective prediction task, the top-ranking prediction precision of the two models reaches 0.0798 and 0.0684 respectively, improvements of 145% and 110% over the mechanism-based Proximity, 40% and 20% over the data-driven method node2vec, and 53% and 31% over the data-driven method edge2vec respectively. Furthermore, applying the Rank Aggregation method to integrate the prediction results of both models leads to precision improvements of 75% and 105%, further enhancing the predictive ability of the models. The prediction results on real-world clinical data of the two models demonstrate sound prediction performance, highlighting their potential to promote the effective application of TCM in novel indications.
As an important therapeutic resource, traditional Chinese medicine (TCM) has undergone thousands of years of clinical practice and application. To promote the modernization of TCM and explore its application potential in new indications, this study draws on research experience from drug repurposing in Western medicine and combines emerging network medicine theories to propose two random walk-based models for predicting potential therapeutic associations between TCM and symptoms: M-RW and GO-DREAMwalk. The two models incorporate path-based and functional information between TCM and symptoms to guide the random walk process. The resulting node sequences are input into a heterogeneous Skip-gram model to learn the embedded vector representations of nodes. Subsequently, an XGBoost classifier is trained by adopting TCM-symptom association labels and the learned embedded vectors. Finally, the models are tested and evaluated by employing clinical data on liver cirrhosis. In the clinically effective prediction task, the top-ranking prediction precision of the two models reaches 0.0798 and 0.0684 respectively, improvements of 145% and 110% over the mechanism-based Proximity, 40% and 20% over the data-driven method node2vec, and 53% and 31% over the data-driven method edge2vec respectively. Furthermore, applying the Rank Aggregation method to integrate the prediction results of both models leads to precision improvements of 75% and 105%, further enhancing the predictive ability of the models. The prediction results on real-world clinical data of the two models demonstrate sound prediction performance, highlighting their potential to promote the effective application of TCM in novel indications.
Available online: November 11, 2025
,DOI: 10.15888/j.cnki.csa.010061
Abstract:
Skin cancer is a common and serious type of cancer, with melanoma having the highest fatality rate. Early detection and treatment can significantly improve the survival rate of skin cancer patients. Dermoscopic, macroscopic, and histopathological images all play essential roles in diagnosis. The application of artificial intelligence technology can effectively enhance the efficiency of classifying these three types of images and help reduce diagnostic costs. Deep learning, with its feature extraction capabilities, is more suitable for the classification tasks of detailed skin cancer images. This study reviews the relevant research on the classification tasks of the three commonly used images in skin cancer diagnosis, analyzes the technical focuses of the three types of images due to their different image characteristics, and conducts targeted analysis of the difficulties faced in clinical application. Finally, future developments and challenges are discussed to promote the broader application of artificial intelligence in skin cancer diagnosis.
Skin cancer is a common and serious type of cancer, with melanoma having the highest fatality rate. Early detection and treatment can significantly improve the survival rate of skin cancer patients. Dermoscopic, macroscopic, and histopathological images all play essential roles in diagnosis. The application of artificial intelligence technology can effectively enhance the efficiency of classifying these three types of images and help reduce diagnostic costs. Deep learning, with its feature extraction capabilities, is more suitable for the classification tasks of detailed skin cancer images. This study reviews the relevant research on the classification tasks of the three commonly used images in skin cancer diagnosis, analyzes the technical focuses of the three types of images due to their different image characteristics, and conducts targeted analysis of the difficulties faced in clinical application. Finally, future developments and challenges are discussed to promote the broader application of artificial intelligence in skin cancer diagnosis.
Available online: November 04, 2025
,DOI: 10.15888/j.cnki.csa.010045
Abstract:
Existing multimodal-based image anomaly detection methods suffer from several limitations: anomaly smoothing during anomaly region extraction, insufficient fine-grained perception, and low discrimination efficiency in defect detection, leading to degraded overall performance. To address these issues, this study proposes a multimodal image anomaly detection model with an asymmetric teacher-student network (MATS), comprising three key components: a cross-modal anomaly amplifier (CAA), a multi-dilated local attention (MDLA) module, and a FastKAN feed-forward network. First, the CAA amplifies anomalous regions while reducing noise by expanding/compressing auxiliary features and fusing them with target features, thus alleviating anomaly smoothing in subsequent detection. Subsequently, the MDLA module enhances fine-grained perception of anomalies through multi-dilation-rate convolutions combined with local attention for multi-scale feature extraction, while integrating normalizing flow (NF) to generate the conditional probability distribution of normal samples. The FastKAN module enables efficient anomaly discrimination via lightweight feature processing, producing feature maps consistent with the teacher network's outputs for pixel-wise distance calculation to evaluate anomaly scores. During testing, regions with significant discrepancies between teacher and student network outputs are identified as anomalies. Experimental results on public industrial image datasets MVTec AD and MVTec 3D-AD demonstrate that the proposed method achieves state-of-the-art performance in multimodal anomaly detection and localization.
Existing multimodal-based image anomaly detection methods suffer from several limitations: anomaly smoothing during anomaly region extraction, insufficient fine-grained perception, and low discrimination efficiency in defect detection, leading to degraded overall performance. To address these issues, this study proposes a multimodal image anomaly detection model with an asymmetric teacher-student network (MATS), comprising three key components: a cross-modal anomaly amplifier (CAA), a multi-dilated local attention (MDLA) module, and a FastKAN feed-forward network. First, the CAA amplifies anomalous regions while reducing noise by expanding/compressing auxiliary features and fusing them with target features, thus alleviating anomaly smoothing in subsequent detection. Subsequently, the MDLA module enhances fine-grained perception of anomalies through multi-dilation-rate convolutions combined with local attention for multi-scale feature extraction, while integrating normalizing flow (NF) to generate the conditional probability distribution of normal samples. The FastKAN module enables efficient anomaly discrimination via lightweight feature processing, producing feature maps consistent with the teacher network's outputs for pixel-wise distance calculation to evaluate anomaly scores. During testing, regions with significant discrepancies between teacher and student network outputs are identified as anomalies. Experimental results on public industrial image datasets MVTec AD and MVTec 3D-AD demonstrate that the proposed method achieves state-of-the-art performance in multimodal anomaly detection and localization.
-
全文下载排行(总排行，年度排行，各期排行)
摘要点击排行(总排行，年度排行，各期排行)
Article Search
Search by issue
Select AllDeselectExport
Display Method:
2016,25(8):1-7
,DOI: 10.15888/j.cnki.csa.005283
Abstract:
Since 2006, Deep Neural Network has achieved huge access in the area of Big Data Processing and Artificial Intelligence, such as image/video discriminations and autopilot. And unsupervised learning methods as the methods getting success in the depth neural network pre training play an important role in deep learning. So, this paper attempts to make a brief introduction and analysis of unsupervised learning methods in deep learning, mainly includs two types, Auto-Encoders based on determination theory and Contrastive Divergence for Restrict Boltzmann Machine based on probability theory. Secondly, the applications of the two methods in Deep Learning are introduced. At last a brief summary and prospect of the challenges faced by unsupervised learning methods in Deep Neural Networks are made.
Since 2006, Deep Neural Network has achieved huge access in the area of Big Data Processing and Artificial Intelligence, such as image/video discriminations and autopilot. And unsupervised learning methods as the methods getting success in the depth neural network pre training play an important role in deep learning. So, this paper attempts to make a brief introduction and analysis of unsupervised learning methods in deep learning, mainly includs two types, Auto-Encoders based on determination theory and Contrastive Divergence for Restrict Boltzmann Machine based on probability theory. Secondly, the applications of the two methods in Deep Learning are introduced. At last a brief summary and prospect of the challenges faced by unsupervised learning methods in Deep Neural Networks are made.
2008,17(5):122-126
Abstract:
随着Internet的迅速发展，网络资源越来越丰富，人们如何从网络上抽取信息也变得至关重要，尤其是占网络资源80%的Deep Web信息检索更是人们应该倍加关注的难点问题。为了更好的研究Deep Web爬虫技术，本文对有关Deep Web爬虫的内容进行了全面、详细地介绍。首先对Deep Web爬虫的定义及研究目标进行了阐述，接着介绍了近年来国内外关于Deep Web爬虫的研究进展，并对其加以分析。在此基础上展望了Deep Web爬虫的研究趋势，为下一步的研究奠定了基础。
随着Internet的迅速发展，网络资源越来越丰富，人们如何从网络上抽取信息也变得至关重要，尤其是占网络资源80%的Deep Web信息检索更是人们应该倍加关注的难点问题。为了更好的研究Deep Web爬虫技术，本文对有关Deep Web爬虫的内容进行了全面、详细地介绍。首先对Deep Web爬虫的定义及研究目标进行了阐述，接着介绍了近年来国内外关于Deep Web爬虫的研究进展，并对其加以分析。在此基础上展望了Deep Web爬虫的研究趋势，为下一步的研究奠定了基础。
2022,31(5):1-20
,DOI: 10.15888/j.cnki.csa.008463
Abstract:
Although the deep learning method has made a huge breakthrough in machine learning, it requires a large amount of manual work for data annotation. Limited by labor costs, however, many applications are expected to reason and judge the instance labels that have never been encountered before. For this reason, zero-shot learning (ZSL) came into being. As a natural data structure that represents the connection between things, the graph is currently drawing more and more attention in ZSL. Therefore, this study reviews the methods of graph-based ZSL systematically. Firstly, the definitions of ZSL and graph learning are outlined, and the ideas of existing solutions for ZSL are summarized. Secondly, the current ZSL methods are classified according to different utilization ways of graphs. Thirdly, the evaluation criteria and datasets concerning graph-based ZSL are discussed. Finally, this study also specifies the problems to be solved in further research on graph-based ZSL and predicts the possible directions of its future development.
Although the deep learning method has made a huge breakthrough in machine learning, it requires a large amount of manual work for data annotation. Limited by labor costs, however, many applications are expected to reason and judge the instance labels that have never been encountered before. For this reason, zero-shot learning (ZSL) came into being. As a natural data structure that represents the connection between things, the graph is currently drawing more and more attention in ZSL. Therefore, this study reviews the methods of graph-based ZSL systematically. Firstly, the definitions of ZSL and graph learning are outlined, and the ideas of existing solutions for ZSL are summarized. Secondly, the current ZSL methods are classified according to different utilization ways of graphs. Thirdly, the evaluation criteria and datasets concerning graph-based ZSL are discussed. Finally, this study also specifies the problems to be solved in further research on graph-based ZSL and predicts the possible directions of its future development.
2011,20(11):80-85
Abstract:
Based on study of current video transcoding solutions, we proposed a distributed transcoding system. Video resources are stored in HDFS(Hadoop Distributed File System) and transcoded by MapReduce program using FFMPEG. In this paper, video segmentation strategy on distributed storage and how they affect accessing time are discussed. We also defined metadata of video formats and transcoding parameters. The distributed transcoding framework is proposed on basis of MapReduce programming model. Segmented source videos are transcoding in map tasks and merged into target video in reduce task. Experimental results show that transcoding time is dependent on segmentation size and trascoding cluster size. Compared with single PC, the proposed distributed video transcoding system implemented on 8 PCs can decrease about 80% of the transcoding time.
Based on study of current video transcoding solutions, we proposed a distributed transcoding system. Video resources are stored in HDFS(Hadoop Distributed File System) and transcoded by MapReduce program using FFMPEG. In this paper, video segmentation strategy on distributed storage and how they affect accessing time are discussed. We also defined metadata of video formats and transcoding parameters. The distributed transcoding framework is proposed on basis of MapReduce programming model. Segmented source videos are transcoding in map tasks and merged into target video in reduce task. Experimental results show that transcoding time is dependent on segmentation size and trascoding cluster size. Compared with single PC, the proposed distributed video transcoding system implemented on 8 PCs can decrease about 80% of the transcoding time.
2012,21(3):260-264
Abstract:
The essential problem of open platform is the validation and authorization of users. Nowadays, OAuth is the international authorization method. Its characteristic is that users could apply to visit their protected resources without the need to enter their names and passwords in the third application. The latest version of OAuth is OAuth2.0 and its operation of validation and authorization are simpler and safer. This paper investigates the principle of OAuth2.0, analyzes the procedure of Refresh Token and offers a design proposal of OAuth2.0 server and specific application examples.
The essential problem of open platform is the validation and authorization of users. Nowadays, OAuth is the international authorization method. Its characteristic is that users could apply to visit their protected resources without the need to enter their names and passwords in the third application. The latest version of OAuth is OAuth2.0 and its operation of validation and authorization are simpler and safer. This paper investigates the principle of OAuth2.0, analyzes the procedure of Refresh Token and offers a design proposal of OAuth2.0 server and specific application examples.
2019,28(6):1-12
,DOI: 10.15888/j.cnki.csa.006915
Abstract:
A knowledge graph is a knowledge base that represents objective concepts/entities and their relationships in the form of graph, which is one of the fundamental technologies for intelligent services such as semantic retrieval, intelligent answering, decision support, etc. Currently, the connotation of knowledge graph is not clear enough and the usage/reuse rate of existing knowledge graphs is relatively low due to lack of documentation. This paper clarifies the concept of knowledge graph through differentiating it from related concepts such as ontology in that the ontology is the schema layer and the logical basis of a knowledge graph while the knowledge graph is the instantiation of an ontology. Research results of ontologies can be used as the foundation of knowledge graph research to promote its developments and applications. Existing generic/domain knowledge graphs are briefly documented and analyzed in terms of building, storage, and retrieval methods. Moreover, future research directions are pointed out.
A knowledge graph is a knowledge base that represents objective concepts/entities and their relationships in the form of graph, which is one of the fundamental technologies for intelligent services such as semantic retrieval, intelligent answering, decision support, etc. Currently, the connotation of knowledge graph is not clear enough and the usage/reuse rate of existing knowledge graphs is relatively low due to lack of documentation. This paper clarifies the concept of knowledge graph through differentiating it from related concepts such as ontology in that the ontology is the schema layer and the logical basis of a knowledge graph while the knowledge graph is the instantiation of an ontology. Research results of ontologies can be used as the foundation of knowledge graph research to promote its developments and applications. Existing generic/domain knowledge graphs are briefly documented and analyzed in terms of building, storage, and retrieval methods. Moreover, future research directions are pointed out.
2002,11(10):70-73
Abstract:
2007,16(9):22-25
Abstract:
本文结合物流遗留系统的实际安全状态,分析了面向对象的编程思想在横切关注点和核心关注点处理上的不足,指出面向方面的编程思想解决方案对系统进行分离关注点处理的优势,并对面向方面的编程的一种具体实现AspectJ进行分析,提出了一种依据AspectJ对遗留物流系统进行IC卡安全进化的方法.
本文结合物流遗留系统的实际安全状态,分析了面向对象的编程思想在横切关注点和核心关注点处理上的不足,指出面向方面的编程思想解决方案对系统进行分离关注点处理的优势,并对面向方面的编程的一种具体实现AspectJ进行分析,提出了一种依据AspectJ对遗留物流系统进行IC卡安全进化的方法.
2011,20(7):184-187,120
Abstract:
According to the actual needs of intelligent household, environmental monitoring etc, this paper designed a wireless sensor node of long-distance communication system. This system used the second SoC CC2530 set in RF and controller chips as the core module and externally connected with CC2591 RF front-end power amplifier module. Based on ZigBee2006 in software agreement stack, it realized each application layer function based on ZStack. It also introduced wireless data acquisition networks based on the ZigBee agreement construction, and has given the hardware design schematic diagram and the software flow chart of sensor node, synchronizer node. The experiment proved that the node is good in performance and the communication is reliable. The communication distance has increased obviously compared with the first generation TI product.
According to the actual needs of intelligent household, environmental monitoring etc, this paper designed a wireless sensor node of long-distance communication system. This system used the second SoC CC2530 set in RF and controller chips as the core module and externally connected with CC2591 RF front-end power amplifier module. Based on ZigBee2006 in software agreement stack, it realized each application layer function based on ZStack. It also introduced wireless data acquisition networks based on the ZigBee agreement construction, and has given the hardware design schematic diagram and the software flow chart of sensor node, synchronizer node. The experiment proved that the node is good in performance and the communication is reliable. The communication distance has increased obviously compared with the first generation TI product.
2016,25(7):8-16
,DOI: 10.15888/j.cnki.csa.005241
Abstract:
This software is completed by Visual c + + 6.0 and Access 2003 tools, and designed in the Unicode character set patterns, to solve the problem about system compatibility and character output garbled in current national language in software development. This development model is used simply, has stable operation, flexible interface, and is convenient for user unified processing (backup, print) vocabulary and voice database at the same time also provides technical guidance to other national language text translation software development. Currently translation supporting tools for Dai region has not yet been released, Daile Wen - Chinese Translation Audible Electronic Dictionary is an important “application innovation” in the field of Dai information technology, it's the basic support of research about minority language cultural information element representation and extraction, and the main function is responsible for Dai queries, translation, reading, etc. Daile Wen - Chinese Translation Electronic Dictionary designed to achieve the common functions such as Dai-Chinese bilingual translation, Dai people reading, Dai phonetic display, it also supports the lexicon to add, modify, delete custom actions, it implements the good human-computer interaction function.
This software is completed by Visual c + + 6.0 and Access 2003 tools, and designed in the Unicode character set patterns, to solve the problem about system compatibility and character output garbled in current national language in software development. This development model is used simply, has stable operation, flexible interface, and is convenient for user unified processing (backup, print) vocabulary and voice database at the same time also provides technical guidance to other national language text translation software development. Currently translation supporting tools for Dai region has not yet been released, Daile Wen - Chinese Translation Audible Electronic Dictionary is an important “application innovation” in the field of Dai information technology, it's the basic support of research about minority language cultural information element representation and extraction, and the main function is responsible for Dai queries, translation, reading, etc. Daile Wen - Chinese Translation Electronic Dictionary designed to achieve the common functions such as Dai-Chinese bilingual translation, Dai people reading, Dai phonetic display, it also supports the lexicon to add, modify, delete custom actions, it implements the good human-computer interaction function.
2008,17(1):113-116
Abstract:
排序是计算机程序设计中一种重要操作，本文论述了C语言中快速排序算法的改进，即快速排序与直接插入排序算法相结合的实现过程。在C语言程序设计中，实现大量的内部排序应用时，所寻求的目的就是找到一个简单、有效、快捷的算法。本文着重阐述快速排序的改进与提高过程，从基本的性能特征到基本的算法改进，通过不断的分析，实验，最后得出最佳的改进算法。
排序是计算机程序设计中一种重要操作，本文论述了C语言中快速排序算法的改进，即快速排序与直接插入排序算法相结合的实现过程。在C语言程序设计中，实现大量的内部排序应用时，所寻求的目的就是找到一个简单、有效、快捷的算法。本文着重阐述快速排序的改进与提高过程，从基本的性能特征到基本的算法改进，通过不断的分析，实验，最后得出最佳的改进算法。
2008,17(8):87-89
Abstract:
随着面向对象软件开发技术的广泛应用和软件测试自动化的要求，基于模型的软件测试逐渐得到了软件开发人员和软件测试人员的认可和接受。基于模型的软件测试是软件编码阶段的主要测试方法之一，具有测试效率高、排除逻辑复杂故障测试效果好等特点。但是误报、漏报和故障机理有待进一步研究。对主要的测试模型进行了分析和分类，同时，对故障密度等参数进行了初步的分析；最后，提出了一种基于模型的软件测试流程。
随着面向对象软件开发技术的广泛应用和软件测试自动化的要求，基于模型的软件测试逐渐得到了软件开发人员和软件测试人员的认可和接受。基于模型的软件测试是软件编码阶段的主要测试方法之一，具有测试效率高、排除逻辑复杂故障测试效果好等特点。但是误报、漏报和故障机理有待进一步研究。对主要的测试模型进行了分析和分类，同时，对故障密度等参数进行了初步的分析；最后，提出了一种基于模型的软件测试流程。
2008,17(8):2-5
Abstract:
本文介绍了一个企业信息门户中单点登录系统的设计与实现。系统实现了一个基于Java EE架构的结合凭证加密和Web Services的单点登录系统，对门户用户进行统一认证和访问控制。论文详细阐述了该系统的总体结构、设计思想、工作原理和具体实现方案，目前系统已在部分省市的广电行业信息门户平台中得到了良好的应用。
本文介绍了一个企业信息门户中单点登录系统的设计与实现。系统实现了一个基于Java EE架构的结合凭证加密和Web Services的单点登录系统，对门户用户进行统一认证和访问控制。论文详细阐述了该系统的总体结构、设计思想、工作原理和具体实现方案，目前系统已在部分省市的广电行业信息门户平台中得到了良好的应用。
2004,13(8):58-59
Abstract:
本文介绍了Visual C++6.0在对话框的多个文本框之间,通过回车键转移焦点的几种方法,并提出了一个改进方法.
本文介绍了Visual C++6.0在对话框的多个文本框之间,通过回车键转移焦点的几种方法,并提出了一个改进方法.
-
全文下载排行(总排行，年度排行，各期排行)
摘要点击排行(总排行，年度排行，各期排行)
Article Search
Search by issue
Select AllDeselectExport
Display Method:
2007,16(10):48-51
Abstract:
论文对HDF数据格式和函数库进行研究,重点以栅格图像为例,详细论述如何利用VC++.net和VC#.net对光栅数据进行读取与处理,然后根据所得到的象素矩阵用描点法显示图像.论文是以国家气象中心开发Micaps3.0(气象信息综合分析处理系统)的课题研究为背景的.
论文对HDF数据格式和函数库进行研究,重点以栅格图像为例,详细论述如何利用VC++.net和VC#.net对光栅数据进行读取与处理,然后根据所得到的象素矩阵用描点法显示图像.论文是以国家气象中心开发Micaps3.0(气象信息综合分析处理系统)的课题研究为背景的.
2002,11(12):67-68
Abstract:
本文介绍非实时操作系统Windows 2000下,利用VisualC++6.0开发实时数据采集的方法.所用到的数据采集卡是研华的PCL-818L.借助数据采集卡PCL-818L的DLLs中的API函数,提出三种实现高速实时数据采集的方法及优缺点.
本文介绍非实时操作系统Windows 2000下,利用VisualC++6.0开发实时数据采集的方法.所用到的数据采集卡是研华的PCL-818L.借助数据采集卡PCL-818L的DLLs中的API函数,提出三种实现高速实时数据采集的方法及优缺点.
2008,17(1):113-116
Abstract:
排序是计算机程序设计中一种重要操作，本文论述了C语言中快速排序算法的改进，即快速排序与直接插入排序算法相结合的实现过程。在C语言程序设计中，实现大量的内部排序应用时，所寻求的目的就是找到一个简单、有效、快捷的算法。本文着重阐述快速排序的改进与提高过程，从基本的性能特征到基本的算法改进，通过不断的分析，实验，最后得出最佳的改进算法。
排序是计算机程序设计中一种重要操作，本文论述了C语言中快速排序算法的改进，即快速排序与直接插入排序算法相结合的实现过程。在C语言程序设计中，实现大量的内部排序应用时，所寻求的目的就是找到一个简单、有效、快捷的算法。本文着重阐述快速排序的改进与提高过程，从基本的性能特征到基本的算法改进，通过不断的分析，实验，最后得出最佳的改进算法。
2008,17(5):122-126
Abstract:
随着Internet的迅速发展，网络资源越来越丰富，人们如何从网络上抽取信息也变得至关重要，尤其是占网络资源80%的Deep Web信息检索更是人们应该倍加关注的难点问题。为了更好的研究Deep Web爬虫技术，本文对有关Deep Web爬虫的内容进行了全面、详细地介绍。首先对Deep Web爬虫的定义及研究目标进行了阐述，接着介绍了近年来国内外关于Deep Web爬虫的研究进展，并对其加以分析。在此基础上展望了Deep Web爬虫的研究趋势，为下一步的研究奠定了基础。
随着Internet的迅速发展，网络资源越来越丰富，人们如何从网络上抽取信息也变得至关重要，尤其是占网络资源80%的Deep Web信息检索更是人们应该倍加关注的难点问题。为了更好的研究Deep Web爬虫技术，本文对有关Deep Web爬虫的内容进行了全面、详细地介绍。首先对Deep Web爬虫的定义及研究目标进行了阐述，接着介绍了近年来国内外关于Deep Web爬虫的研究进展，并对其加以分析。在此基础上展望了Deep Web爬虫的研究趋势，为下一步的研究奠定了基础。
2026,35(1):1-18, DOI: 10.15888/j.cnki.csa.010074
Abstract:
2026,35(1):39-51, DOI: 10.15888/j.cnki.csa.010055
Abstract:
2026,35(1):19-38, DOI: 10.15888/j.cnki.csa.010029
Abstract:
2025,34(12):1-15, DOI: 10.15888/j.cnki.csa.010033
Abstract:
2025,34(11):1-19, DOI: 10.15888/j.cnki.csa.010002
Abstract:
2025,34(10):16-31, DOI: 10.15888/j.cnki.csa.009979
Abstract:
2025,34(10):1-15, DOI: 10.15888/j.cnki.csa.009975
Abstract:
2025,34(8):1-13, DOI: 10.15888/j.cnki.csa.009907
Abstract:
2025,34(7):1-13, DOI: 10.15888/j.cnki.csa.009942
Abstract:
2025,34(6):1-11, DOI: 10.15888/j.cnki.csa.009903
Abstract:
2025,34(5):1-13, DOI: 10.15888/j.cnki.csa.009879
Abstract:
2025,34(4):18-33, DOI: 10.15888/j.cnki.csa.009865
Abstract:
2025,34(4):1-17, DOI: 10.15888/j.cnki.csa.009839
Abstract:
2025,34(1):1-10, DOI: 10.15888/j.cnki.csa.009782
Abstract:
2025,34(1):26-36, DOI: 10.15888/j.cnki.csa.009733
Abstract:
2025,34(1):11-25, DOI: 10.15888/j.cnki.csa.009709
Abstract:
2024,33(12):1-15, DOI: 10.15888/j.cnki.csa.009708
Abstract:
2024,33(11):38-47, DOI: 10.15888/j.cnki.csa.009702
Abstract:
2024,33(10):26-36, DOI: 10.15888/j.cnki.csa.009646
Abstract:
2024,33(6):16-27, DOI: 10.15888/j.cnki.csa.009559
Abstract:
2024,33(6):1-15, DOI: 10.15888/j.cnki.csa.009531
Abstract:
2024,33(4):50-59, DOI: 10.15888/j.cnki.csa.009486
Abstract:
Application Review of Convolutional Neural Networks in Pathological Images Diagnosis of Liver Cancer
2024,33(4):26-38, DOI: 10.15888/j.cnki.csa.009466
Abstract:
2024,33(3):1-11, DOI: 10.15888/j.cnki.csa.009433
Abstract:
2024,33(3):12-23, DOI: 10.15888/j.cnki.csa.009421
Abstract:
2024,33(1):37-48, DOI: 10.15888/j.cnki.csa.009377
Abstract:
2024,33(1):22-36, DOI: 10.15888/j.cnki.csa.009369
Abstract:
2023,32(12):32-42, DOI: 10.15888/j.cnki.csa.009356
Abstract:
2023,32(11):36-47, DOI: 10.15888/j.cnki.csa.009269
Abstract:
2023,32(9):1-18, DOI: 10.15888/j.cnki.csa.009226
Abstract:
2023,32(8):1-18, DOI: 10.15888/j.cnki.csa.009208
Abstract:
2023,32(4):1-15, DOI: 10.15888/j.cnki.csa.009038
Abstract:
2023,32(4):16-41, DOI: 10.15888/j.cnki.csa.009037
Abstract:
2023,32(3):25-47, DOI: 10.15888/j.cnki.csa.008980
Abstract:
2023,32(3):1-14, DOI: 10.15888/j.cnki.csa.008976
Abstract:
2023,32(2):1-12, DOI: 10.15888/j.cnki.csa.008743
Abstract:
2023,32(1):12-28, DOI: 10.15888/j.cnki.csa.008914
Abstract:
2023,32(1):29-40, DOI: 10.15888/j.cnki.csa.008909
Abstract:
2023,32(1):1-11, DOI: 10.15888/j.cnki.csa.008848
Abstract:
2023,32(1):41-49, DOI: 10.15888/j.cnki.csa.008805
Abstract:
2022,31(12):20-28, DOI: 10.15888/j.cnki.csa.008838
Abstract:
2022,31(12):1-9, DOI: 10.15888/j.cnki.csa.008825
Abstract:
2022,31(11):49-59, DOI: 10.15888/j.cnki.csa.008847
Abstract:
2022,31(11):31-48, DOI: 10.15888/j.cnki.csa.008811
Abstract:
2022,31(10):44-50, DOI: 10.15888/j.cnki.csa.008776
Abstract:
2022,31(10):25-35, DOI: 10.15888/j.cnki.csa.008734
Abstract:
2022,31(10):36-43, DOI: 10.15888/j.cnki.csa.008711
Abstract:
2022,31(9):57-69, DOI: 10.15888/j.cnki.csa.008661
Abstract:
2022,31(8):1-16, DOI: 10.15888/j.cnki.csa.008713
Abstract:
2022,31(7):12-22, DOI: 10.15888/j.cnki.csa.008641
Abstract:
2022,31(7):35-45, DOI: 10.15888/j.cnki.csa.008595
Abstract:
2022,31(7):46-54, DOI: 10.15888/j.cnki.csa.008590
Abstract:
2022,31(7):23-34, DOI: 10.15888/j.cnki.csa.008566
Abstract:
2022,31(7):1-11, DOI: 10.15888/j.cnki.csa.008545
Abstract:
2022,31(5):40-51, DOI: 10.15888/j.cnki.csa.008476
Abstract:
2022,31(5):1-20, DOI: 10.15888/j.cnki.csa.008463
Abstract:
2022,31(5):21-29, DOI: 10.15888/j.cnki.csa.008426
Abstract:
2022,31(4):14-32, DOI: 10.15888/j.cnki.csa.008419
Abstract:
2022,31(4):1-13, DOI: 10.15888/j.cnki.csa.008418
Abstract:
2022,31(4):47-58, DOI: 10.15888/j.cnki.csa.008403
Abstract:
2022,31(3):1-8, DOI: 10.15888/j.cnki.csa.008415
Abstract:
2022,31(3):48-55, DOI: 10.15888/j.cnki.csa.008377
Abstract:
2022,31(2):13-21, DOI: 10.15888/j.cnki.csa.008319
Abstract:
2022,31(2):1-12, DOI: 10.15888/j.cnki.csa.008303
Abstract:
2022,31(1):1-10, DOI: 10.15888/j.cnki.csa.008323
Abstract:
2021,30(12):28-36, DOI: 10.15888/j.cnki.csa.008289
Abstract:
2021,30(12):10-17, DOI: 10.15888/j.cnki.csa.008252
Abstract:
2021,30(12):1-9, DOI: 10.15888/j.cnki.csa.008245
Abstract:
2021,30(10):21-30, DOI: 10.15888/j.cnki.csa.008137
Abstract:
2021,30(9):12-23, DOI: 10.15888/j.cnki.csa.008086
Abstract:
2021,30(9):24-40, DOI: 10.15888/j.cnki.csa.008079
Abstract:
2021,30(9):41-52, DOI: 10.15888/j.cnki.csa.008078
Abstract:
2021,30(8):14-21, DOI: 10.15888/j.cnki.csa.008003
Abstract:
2021,30(7):13-21, DOI: 10.15888/j.cnki.csa.007992
Abstract:
2021,30(7):1-12, DOI: 10.15888/j.cnki.csa.007973
Abstract:
2021,30(6):18-27, DOI: 10.15888/j.cnki.csa.007956
Abstract:
2021,30(5):21-30, DOI: 10.15888/j.cnki.csa.007916
Abstract:
2021,30(3):1-13, DOI: 10.15888/j.cnki.csa.007839
Abstract:
2021,30(1):10-18, DOI: 10.15888/j.cnki.csa.007776
Abstract:
2020,29(12):13-25, DOI: 10.15888/j.cnki.csa.007701
Abstract:
2020,29(11):29-39, DOI: 10.15888/j.cnki.csa.007705
Abstract:
2020,29(9):16-25, DOI: 10.15888/j.cnki.csa.007632
Abstract:
2020,29(9):26-31, DOI: 10.15888/j.cnki.csa.007617
Abstract:
2020,29(9):1-15, DOI: 10.15888/j.cnki.csa.007489
Abstract:
2020,29(8):1-8, DOI: 10.15888/j.cnki.csa.007478
Abstract:
2020,29(6):56-63, DOI: 10.15888/j.cnki.csa.007460
Abstract:
2020,29(6):1-12, DOI: 10.15888/j.cnki.csa.007431
Abstract:
2020,29(6):13-21, DOI: 10.15888/j.cnki.csa.007430
Abstract:
2020,29(3):29-38, DOI: 10.15888/j.cnki.csa.007315
Abstract:
2020,29(3):1-10, DOI: 10.15888/j.cnki.csa.007301
Abstract:
2020,29(3):11-19, DOI: 10.15888/j.cnki.csa.007261
Abstract:
2020,29(2):1-8, DOI: 10.15888/j.cnki.csa.007275
Abstract:
2020,29(2):9-27, DOI: 10.15888/j.cnki.csa.007262
Abstract:
2020,29(1):14-21, DOI: 10.15888/j.cnki.csa.007243
Abstract:
2020,29(1):22-28, DOI: 10.15888/j.cnki.csa.007237
Abstract:
2019,28(12):1-8, DOI: 10.15888/j.cnki.csa.007195
Abstract:
2019,28(11):1-9, DOI: 10.15888/j.cnki.csa.007156
Abstract:
2019,28(9):1-8, DOI: 10.15888/j.cnki.csa.007037
Abstract:
2019,28(7):1-8, DOI: 10.15888/j.cnki.csa.006988
Abstract:
2019,28(6):1-12, DOI: 10.15888/j.cnki.csa.006915
Abstract:
2019,28(4):205-211, DOI: 10.15888/j.cnki.csa.006739
Abstract:
2019,28(3):1-9, DOI: 10.15888/j.cnki.csa.006812
Abstract:
2019,28(3):10-17, DOI: 10.15888/j.cnki.csa.006806
Abstract:
2019,28(1):1-9, DOI: 10.15888/j.cnki.csa.006720
Abstract:
2019,28(1):10-16, DOI: 10.15888/j.cnki.csa.006707
Abstract:
2018,27(10):39-45, DOI: 10.15888/j.cnki.csa.006591
Abstract:
2018,27(10):33-38, DOI: 10.15888/j.cnki.csa.006560
Abstract:
2018,27(9):47-51, DOI: 10.15888/j.cnki.csa.006542
Abstract:
2018,27(8):10-18, DOI: 10.15888/j.cnki.csa.006456
Abstract:
2018,27(7):1-10, DOI: 10.15888/j.cnki.csa.006451
Abstract:
2018,27(5):1-9, DOI: 10.15888/j.cnki.csa.006336
Abstract:
2018,27(4):1-9, DOI: 10.15888/j.cnki.csa.006273
Abstract:
2018,27(1):1-9, DOI: 10.15888/j.cnki.csa.006169
Abstract:
2017,26(12):32-36, DOI: 10.15888/j.cnki.csa.006129
Abstract:
2017,26(12):25-31, DOI: 10.15888/j.cnki.csa.006096
Abstract:
2017,26(6):36-39, DOI: 10.15888/j.cnki.csa.005807
Abstract:
2017,26(4):1-8, DOI: 10.15888/j.cnki.csa.005773
Abstract:
2017,26(2):30-36, DOI: 10.15888/j.cnki.csa.005626
Abstract:
2016,25(9):17-26, DOI: 10.15888/j.cnki.csa.005354
Abstract:
2016,25(9):10-16, DOI: 10.15888/j.cnki.csa.005330
Abstract:
2016,25(8):1-7, DOI: 10.15888/j.cnki.csa.005283
Abstract:
2016,25(6):9-16, DOI: 10.15888/j.cnki.csa.005173
Abstract:
2016,25(6):1-8, DOI: 10.15888/j.cnki.csa.005206
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
下载归智APP ，关注本刊
External Links
You are the firstVisitors
Copyright: Institute of Software, Chinese Academy of Sciences Beijing ICP No. 05046678-3
Address：4# South Fourth Street, Zhongguancun,Haidian, Beijing,Postal Code：100190
Phone：010-62661041 Fax： Email：csa (a) iscas.ac.cn
Technical Support：Beijing Qinyun Technology Development Co., Ltd.
Copyright: Institute of Software, Chinese Academy of Sciences Beijing ICP No. 05046678-3
Address：4# South Fourth Street, Zhongguancun,Haidian, Beijing,Postal Code：100190
Phone：010-62661041 Fax： Email：csa (a) iscas.ac.cn
Technical Support：Beijing Qinyun Technology Development Co., Ltd.
