• Volume 30,Issue 3,2021 Table of Contents
    Select All
    Display Type: |
    • >Survey
    • Overview on Deep Learning-Based Object Detection

      2021, 30(3):1-13. DOI: 10.15888/j.cnki.csa.007839

      Abstract (1437) HTML (7601) PDF 2.51 M (3562) Comment (0) Favorites

      Abstract:Object detection is a research hotspot in the field of computer vision. In recent years, the deep learning algorithms contributing to object detection has developed by leaps and bounds. Objection detection algorithms based on deep learning can be roughly divided into two categories depending on candidate regions and regression, respectively. The object detection algorithms based on candidate regions have high accuracy, but complex structure and low speed of detection. The object detection algorithms based on regression, contrarily, have simple structure, high speed of detection, and thus more applications in the field of real-time object detection, but its detection is with low accuracy. This paper summarizes the mainstream algorithms of object detection based on deep learning and analyzes the advantages and disadvantages of different algorithms and their applications. Finally, this paper predicts the prospects of deep learning-based object detection algorithms according to the existing challenges.

    • Survey of Speech Recognition and End-to-End Techniques

      2021, 30(3):14-23. DOI: 10.15888/j.cnki.csa.007852

      Abstract (1078) HTML (4064) PDF 1.35 M (2534) Comment (0) Favorites

      Abstract:The paper briefly introduces the history and application of speech recognition, traditional speech recognition techniques, and current research progress. Traditional speech recognition relies on statistics-based methods and sound spectrum features to train Gaussian Mixture Model-Hidden Markov Model (GMM-HMM) hybrid model. Nowadays, speech recognition models are mainly based on deep learning. Generally, Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) can effectively extract features to establish acoustic models. Further research depends on end-to-end techniques to avoid error transmission among models, and these techniques mainly include Connectionist Temporal Classification (CTC) and attention. The latest models and methods highlight attention, which are trying to integrate it with CTC to achieve better results. Finally, combined with the authors’ understanding, the paper summarizes the existing problems and future development in speech recognition.

    • Graph Network with Dependency Constraints for Joint Entity and Relationship Extraction

      2021, 30(3):24-32. DOI: 10.15888/j.cnki.csa.007825

      Abstract (786) HTML (1116) PDF 1.39 M (1984) Comment (0) Favorites

      Abstract:Entity relationship extraction is one of the key tasks of information extraction, which involves a multi-task cascade including entity extraction and relationship extraction. Traditional methods of entity relationship extraction follow a mode of Pipeline which separates entity extraction from relationship extraction, ignoring the internal connection between the two. As a result, the effect of relationship extraction depends heavily on entity extraction, and it is prone to error accumulation. To avoid this problem, we propose an end-to-end joint entity and relationship extraction model, which relies on the self-attention mechanism to learn word features, constructs dependency constraints based on dependency information contained in syntactic dependency graphs, and then integrates constraint information into a graph attention network for entity and relationship extraction. Experiments on the public data set NYT demonstrate the advance and significance of our model which has a high recall rate and better extraction performance than previous methods.

    • Data Asset Management System Based on Data Center

      2021, 30(3):33-42. DOI: 10.15888/j.cnki.csa.007836

      Abstract (959) HTML (5979) PDF 2.25 M (2430) Comment (0) Favorites

      Abstract:After more than ten years of informatization construction, the company’s information system has fully covered businesses and applications at all levels, such as business operations, grid operations, and customer service, providing strong support for the efficient operation of businesses. However, a unified enterprise-level data asset management system is still lacking. To further raise the value of data in power grid enterprises, this study proposes a set of enterprise-level data asset management system based on the data center and discusses data quality optimization, data asset management, and data sharing service.

    • Multi-Band Satellite Data Fusion Based on Two Improved Wavelet Algorithms

      2021, 30(3):43-51. DOI: 10.15888/j.cnki.csa.007837

      Abstract (562) HTML (1021) PDF 1.85 M (1870) Comment (0) Favorites

      Abstract:Two improved fusion algorithms have been constituted, wavelet fusion based on HSV color model (HSV-WT) and an Adaptive Wavelet Packet (AWP) based on region features, which were applied to processing satellite data, MultiSpectral (MS) LandSat TM & Panchromatic (P) SPOT-5, and LandSat TM MS & Synthetic Aperture Radar (SAR). The proposed HSV-WT & AWP algorithms enhance the fused image’s ability to express the spatial details while preserving spectral information of the MS data. Experimental results demonstrate that AWP performs better than HSV-WT in fusing the same data. These two algorithms, compared with traditional wavelet algorithms, can help to overcome defects when processing high-frequency signals, and they are appropriate for fusing data if the ratios of spatial resolution between the two images to be fused are not in 2n relationships.

    • Yellowing Defect Detection of Vehicle Light Guide Plate Based on Color Perception

      2021, 30(3):52-59. DOI: 10.15888/j.cnki.csa.007813

      Abstract (650) HTML (853) PDF 1.59 M (1677) Comment (0) Favorites

      Abstract:Yellow defects are inevitable due to the high production temperature of the fixture to prepare light guide plates (LGPs). This study proposes a method for detecting yellowing defects of LGP based on machine vision. Firstly, a bilateral filter is designed after gray-level transformation to reduce noise impact. Secondly, the outline of the LGPs is highlighted by the difference of neighbor pixels. Then, the contour extraction and segmentation of three LGPs are completed by the proposed self-adapting threshold filling algorithm and the line segment distance threshold. Finally, according to LGP coordinates, rectangular regions can be generated, and 81-dimensional eigenvectors and a Support Vector Machine (SVM) model can be built. A large number of experiments were carried out on the basis of the LGP images collected in the industrial field. Experimental results prove that the algorithm has high running efficiency and strong robustness and still presents high detection accuracy in the case of few training samples.

    • Hybrid Parallel Streamline Generation System Suitable for Heterogeneous Clusters

      2021, 30(3):60-69. DOI: 10.15888/j.cnki.csa.007842

      Abstract (640) HTML (778) PDF 1.37 M (1331) Comment (0) Favorites

      Abstract:Streamline is one of the main methods of flow visualization. In light of a large amount of computation, the streamline generation from large flow fields usually requires parallel computing environments, such as high-performance computers and parallel algorithms, to accelerate computation. As wider application of heterogeneous computing systems, we design a hybrid parallel streamline generation system suitable for heterogeneous clusters in terms of data decomposition, overlapping and communication strategy with technologies such as data-parallel primitives and message passing interface to maximize the computing power of the heterogeneous parallel computing environment and achieve more efficient parallel streamline generation. A set of algorithms related to parallel particle advection are proposed and implemented. The system is deployed on a domestic supercomputer, and experiments are conducted to visualize the results of a large-scale CFD flow field simulation. This study provides relevant experimental results and analyzes the performance, scalability, and load balance of the core parallel algorithm, verifying the effectiveness and scalability of the system and algorithm.

    • Driver Learning System Based on Qt in Environment of Kylin OS

      2021, 30(3):70-78. DOI: 10.15888/j.cnki.csa.007849

      Abstract (866) HTML (1271) PDF 2.16 M (2062) Comment (0) Favorites

      Abstract:Amid the wave of basic software localization, Kylin Operating System (OS) has been applied to industries. With the powerful C++ API framework provided by Qt, this study develops a kernel driven learning system running on the domestic Kylin OS to make up for the lack of driver learning software. This study mainly focuses on the execution process of hundreds of drivers in the kernel, the general architecture of Linux kernel drivers, the detailed classification of kernel drivers, driver configuration and debugging technology, the relationship among applications, kernel, hardware, drivers, and other driver learning mechanisms. The system is implemented based on modular design with layered software architecture. It enables network request, remote file download/upload, general architecture of kernel driver, detailed driver classification, driver learning mechanism, video learning, specific driver, system settings (including two sub modules of computer system information and system upgrade detection). Finally, it is applied to the computers of X86, ARM, and MIPs through cross compilation.

    • Defect Detection of Motor Cover Based on Improved YOLOv4

      2021, 30(3):79-87. DOI: 10.15888/j.cnki.csa.007808

      Abstract (760) HTML (1162) PDF 1.92 M (1339) Comment (0) Favorites

      Abstract:Adaptive histogram equalization with limited contrast is applied to strengthening the target feature to solve the problem of unclear targets in a complex background during crack detection of motor covers based on machine vision. A systematic dataset construction scheme is proposed by comparing Mosaic and CutMix data augmentation and combining with a variety of data enhancement techniques to address the low generalization of the model induced by the small volume of training data in the machine vision system and single background of training pictures. Besides, a weighted fusion loss function combined with adaptive multi-scale focus loss and CIoU loss is proposed to deal with the low detection rate caused by unbalanced numbers of positive and negative samples in the single class detection and small target detection of YOLOv4, and the optimal hyper parameters are obtained through experiments. Finally, the anchor box is initialized by the K-means algorithm to make the model more suitable for predicting linear targets. Results demonstrate that this method achieves an Average Precision (AP) of 95.8% for detecting crack types, which is 9.7% higher than before, and the single-sheet detection time is 48 ms, presenting the potential for engineering application.

    • X-Ray Detection System for Foreign Bodies in Sole of Shoes Based on Deep Learning

      2021, 30(3):88-94. DOI: 10.15888/j.cnki.csa.007820

      Abstract (638) HTML (719) PDF 1.67 M (1179) Comment (0) Favorites

      Abstract:Broken needles are frequently seen in the production process of clothing and shoe factories. This study proposes a detection system of metal foreign bodies in sole of shoes based on deep learning since those residual bodies such as broken needles in shoes will threaten people’s safety. Firstly, shoes are put on a conveyor belt in turn and sent to a needle detector, and the images are collected by X-ray irradiation. After that, the images are preprocessed to highlight the small metal foreign bodies. Finally, metal foreign bodies and their positions are detected with a deep learning network model. Experimental results show that preprocessing images and fine-tuning the label box can make metal foreign bodies clearer, and the average precision of the model is 97.6%. It proves that the model can effectively detect the metal foreign bodies with different shapes left in footwear, presenting great commercial potential.

    • Comprehensive Evaluation Equipment for Fire Control System of Certain Artillery Fighting Vehicle

      2021, 30(3):95-102. DOI: 10.15888/j.cnki.csa.007826

      Abstract (571) HTML (602) PDF 1.57 M (1105) Comment (0) Favorites

      Abstract:This study systematically discusses the development process of a comprehensive evaluation device for the fire control system of a certain type of artillery combat vehicles and clarifies the establishment of a mathematical model for the accuracy evaluation of each stand-alone device in the fire control system and the specific calculation method. On the basis of network packet capture, numerical analysis and mathematical statistics theory, a comprehensive evaluation instrument for fire control system of certain artillery fighting vehicle is designed, with the functions of packet capture and storage, packet analysis and display, data statistics, and analysis. The equipment can reflect the failure, status and system accuracy of the fire control system through packet analysis, bringing great convenience to the experimental debugging in the fire control system of combat vehicles. It has been applied to the experimental debugging for a certain type of combat vehicles, achieving good performance.

    • Monitoring System of Wheel Hub Assembly Line Based on OPC Technology

      2021, 30(3):103-109. DOI: 10.15888/j.cnki.csa.007832

      Abstract (541) HTML (593) PDF 1.28 M (1029) Comment (0) Favorites

      Abstract:The function of the host computer developed by the high-level programming language in the current production line is undefined and it is difficult to integrate with OPC technology. Therefore, this study combines the operation process of the general production line to design a host computer monitoring system based on OPC technology, which is applied to an actual assembly line. Qt and KEPServerEX are combined to construct an efficient OPC communication mechanism in the system, and the data received from the server is input into modules of storage, real-time display calculation, alarm processing, and chart analysis for further processing. In addition, the system also includes control modules such as station check and marking, enabling the client’s control over the PLC. Experimental results prove that the system has high stability and complete functions, which can meet the requirements of the host computer software in most production lines.

    • Algorithm Integration in Distributed Computing System

      2021, 30(3):110-116. DOI: 10.15888/j.cnki.csa.007804

      Abstract (749) HTML (1523) PDF 1.47 M (1185) Comment (0) Favorites

      Abstract:Algorithm integration is an important content in developing a comprehensive data processing and analysis system. As the development of a near-space science experiment system, a method to integrate algorithms of multiple types, versions, and functions in a large distributed system is designed to facilitate the algorithm integration in a data processing and analysis system. Parallel development, plug-in calling, flexible expansion, and update iteration of the scientific experiment system are enabled by constructing algorithm plug-ins. The system improved with this method is applied to the ZStack private cloud platform, and the method is verified by calculation and analysis of atmospheric parameters, in-situ detection parameters, and ecological parameters in near-space science experiments.

    • Differential Evolution Algorithm Based on Low-Density Individual in Neighborhood for Multimodal Optimization Problem

      2021, 30(3):117-125. DOI: 10.15888/j.cnki.csa.007824

      Abstract (620) HTML (665) PDF 1.31 M (1157) Comment (0) Favorites

      Abstract:A differential evolution algorithm based on low-density individuals in the neighborhood is proposed to solve MultiModal Optimization Problems (MMOPs). In each generation, the algorithm first relies on density peak clustering to find the density of each individual and then take the lower-density individuals in the neighborhood of the current individual as a base vector of the mutation operator. As the population evolves, the algorithm will automatically transform from the exploration stage to the convergence stage, thereby balancing its exploration and convergence capabilities. The proposed algorithm is applied to the CEC2013 multimodal benchmark function for simulation experiments. Results demonstrate that the algorithm has obvious advantages over other multimodal optimization algorithms based on differential evolution in evaluating the peak ratios and stability of indexes, and the advantage is more distinct with the increasing dimensionality and complexity of the test function. It behaves better than many existing multimodal optimization algorithms based on differential evolution.

    • Application of Improved CNN-LSTM Model in Fault Diagnosis of Rolling Bearings

      2021, 30(3):126-133. DOI: 10.15888/j.cnki.csa.007830

      Abstract (792) HTML (3121) PDF 1.27 M (1823) Comment (0) Favorites

      Abstract:The state of rolling bearings has a great influence on the working state of the whole machine, but the fault diagnosis method of the rolling bearings at present has some problems, such as dependency on manual feature extraction and low robustness. Therefore, we propose a fault diagnosis method of rolling bearings (1D-CNN-LSTM) based on the improved integration of 1D Convolutional Neural Network (1D-CNN) and Long Short-Term Memory (LSTM) network. Firstly, the 1D-CNN-LSTM model is used to classify and identify six different working states of rolling bearings. The experimental results indicate that the proposed classification model can identify different states of rolling bearings at a high speed, with an average identification accuracy of 99.83%. Secondly, the proposed model is compared with some traditional algorithm models and shows great advantages in measuring accuracy. Finally, transfer learning is introduced to test the robustness and generalization ability of the proposed model. The experimental results demonstrate that the model proposed in this study has good adaptability and high efficiency under different working conditions, featuring strong generalization ability and engineering application feasibility.

    • Greedy Algorithms Optimized Virtual Machine Allocation for Cloud Data Centers

      2021, 30(3):134-141. DOI: 10.15888/j.cnki.csa.007814

      Abstract (732) HTML (700) PDF 1.03 M (1395) Comment (0) Favorites

      Abstract:It is a critical problem to uniformly allocate a large number of virtual machines at the cloud clients to the physical hosts at the cloud data centers. To this end, a greedy algorithm optimized virtual machine allocation approach for cloud data centers is proposed in this paper. First, a working scenario is designed for the enterprise-oriented cloud data centers, including three layers, a user layer, the layer of cloud service provider, and the star layer of cloud data centers. Specifically, the user layer is used to generate the request sets of the virtual machines, and the layer of cloud service provider allocates a large number of request sets of the virtual machines at the user layer to the bottom cloud data center through the classical bin packing algorithm. Then, the mathematical models considering different constraints are established during the allocation of the virtual machines. Finally, the virtual machine allocation among the cloud data centers is optimized using the greedy algorithm. In addition, the big data center of an enterprise is taken as the cloud testing environment, and the test results show that the classical Best-Fit-Algorithm (BFA) performs well in virtual machine allocation and consumes little energy of cloud platforms, providing a reference for the construction of cloud data centers in other enterprises.

    • Artificial Bee Colony Algorithm in Parameter Estimation of Chaotic Systems

      2021, 30(3):142-150. DOI: 10.15888/j.cnki.csa.007831

      Abstract (575) HTML (579) PDF 2.91 M (957) Comment (0) Favorites

      Abstract:In order to estimate the unknown parameters of the chaotic systems more accurately, we propose a method for the parameter estimation of chaotic systems based on the artificial bee colony algorithm. This method converts the parameter estimation of chaotic systems to the function optimization problem of a multi-dimensional variable and then uses a search equation to fully search the multi-dimensional spatial variable. Furthermore, the optimized artificial bee colony algorithm is applied to calculate the mean square error between the estimated value and the true value, so as to realize the parameter estimation in the chaotic systems. In addition, the results of parameter estimation simulation of the Lorenz chaotic system indicate the feasibility of the proposed method. Besides, the improved algorithm has fast convergence and high estimation accuracy.

    • Application of Improved Generative Adversarial Network in Scene Image Translation

      2021, 30(3):151-157. DOI: 10.15888/j.cnki.csa.007816

      Abstract (591) HTML (904) PDF 1.22 M (1249) Comment (0) Favorites

      Abstract:Aiming at the translation between different scene images, we propose an improved generative adversarial network model that can generate high-quality target scene images. In the process of generating a target image, the spatial position information of the original image will lose due to down sampling. Therefore, a generative network that includes jump connections and residual blocks is designed in this paper. By adding multiple jump connections to the network, we can keep the spatial position information of the image transmitting in the network. At the same time, to improve the stability of the generated image during the training, we introduce the Structural Similarity Index Measure (SSIM) as a structure reconstruction loss to guide the model to generate a target image with a better structure. In addition, in order to make the translated target scene image retain more color details, we add an identity preservation loss, obviously enhancing the color expressiveness of the target generated image. The experimental results show that the improved generative adversarial network model proposed in this study can be effectively applied in scene image translation.

    • Point Cloud Registration Method Based on Kinect

      2021, 30(3):158-163. DOI: 10.15888/j.cnki.csa.007733

      Abstract (741) HTML (1316) PDF 1.08 M (1401) Comment (0) Favorites

      Abstract:The point clouds collected by Kinect have a large quantity and position errors, and it is inefficient to directly apply the Iterative Closest Point (ICP) algorithm to point cloud registration. To solve this problem, we propose an improved point cloud registration algorithm based on the angle between the normal vectors of feature points. First, the voxel grids are used to down sample the original point clouds collected by Kinect and reduce the number of point clouds and a filter is applied to remove the outliers. Then, the Scale Invariant Feature Transform (SIFT) algorithm is employed to extract the common feature points between the target point clouds and the point clouds to be registered, and the angle between the normal vectors of feature points is calculated to adjust the point cloud pose. Thus, the initial registration of the point clouds is completed. Finally, the ICP algorithm is applied to complete the fine registration of the point clouds. The experimental results show that compared with the traditional ICP algorithm, the proposed algorithm, while ensuring the registration accuracy, can improve the registration efficiency of point clouds and has high applicability and robustness.

    • Phone Usage Monitoring Based on Human Pose Estimation

      2021, 30(3):164-170. DOI: 10.15888/j.cnki.csa.007835

      Abstract (715) HTML (1870) PDF 1.36 M (1230) Comment (0) Favorites

      Abstract:The increasing popularity of smartphones brings not only convenience to people but also a lot of hidden dangers. Thus, it is necessary to monitor and restrict the use of phones in some specific situations. In this paper, we design a system to monitor phone usage. First, YOLOv3 is used to detect human bodies in an image. Then, the joints for each person are obtained by the OpenPose pose estimation algorithm. Furthermore, YOLOv3 is employed to judge whether there is a mobile phone in the hands. Finally, the current phone usage status is recognized by a neural network classifier. The experimental results show that the proposed scheme has good detection and recognition performance and can meet the application requirements in relevant scenarios.

    • No-Load Detection of Underground Conveyor Belt in Coal Mine

      2021, 30(3):171-176. DOI: 10.15888/j.cnki.csa.007807

      Abstract (630) HTML (802) PDF 986.32 K (1310) Comment (0) Favorites

      Abstract:The long-time and high-speed running of an underground conveyor belt in a coal mine will consume a lot of electricity in the case of no coal or little coal. In order to reduce the electricity loss caused by the underground conveyor belts, we propose a no-load determination method of the conveyor belts combining the YOLOv3 algorithm with the edge-based structural similarity algorithm. First, the structure features and edge features are fused by the edge-based structural similarity algorithm. Then, the similarity of adjacent 10 frames of images is successively compared three times to judge the running state of a conveyor belt. If the conveyor belt is running, the YOLOv3 model based on the adaptive anchor box mechanism is used to detect the coal amount on the conveyor belt. Finally, whether the conveyor belt carries a load or not is judged. The experimental results show that the proposed method can effectively and accurately judge the no-load state of the conveyor belts and the detection accuracy reaches 96.85%.

    • Text Classification Model Based on Width and Word Vector Feature

      2021, 30(3):177-183. DOI: 10.15888/j.cnki.csa.007827

      Abstract (667) HTML (613) PDF 995.19 K (1219) Comment (0) Favorites

      Abstract:To resolve the issues of weak memory ability and no global word feature information in the word-vector-based text classification model, we propose a text classification model (WideText) based on the width and word vector features. Firstly, text cleaning, word segmentation, unit encoding and dictionary definitions are carried out. Secondly, the Term Frequency-Inverse Document Frequency (TF-IDF) index of the global word units is calculated and each text is vectorized. Furthermore, the words in the input text are mapped to the word embedding matrix through encoding. After the word vector features are embedded and averagely superimposed, they are spliced with the text vector features based on TF-IDF and transmitted to the output layer. Finally, the probability of the features belonging to each category is calculated. The proposed model combines the expressive ability of text vector features on the basis of low-dimensional word vectors and has excellent generalization and memory abilities. The experimental results show that after the introduction of the width feature, the WideText classification performance is significantly improved in comparison with that in the word-vector-based text classification model and also slightly better than that in the feedforward neural network classifiers.

    • Ultra-Short-Term Load Interval Prediction Based on Least Squares Support Vector Machine

      2021, 30(3):184-189. DOI: 10.15888/j.cnki.csa.007805

      Abstract (566) HTML (2161) PDF 1.05 M (1338) Comment (0) Favorites

      Abstract:The rapid development of smart grids has brought new challenges to grid operation. In order to adapt to the requirements of rapid response of smart grids and to rapidly estimate the future operation trend of power loads, we propose a prediction method of an ultra-short-term power load interval based on the least squares support vector machine (LSSVM) model. This method predicts the interval by estimating the overall noise variance of the sample data on the basis of point prediction. which has a small calculated amount and greatly reduces the prediction time consumption. With regard to model parameter selection, the optimal training sample size and embedding dimensions are first determined using the parameter determination method of Gamma Test noise estimation, and then the optimal hyper-parameters are selected by the grid search method so that the fitting error of the LSSVM model on the training samples approximates the estimated minimum noise. To verify the validity of the proposed method in this paper, we apply the scheduling load data from a certain grid.to simulation experiments. The results show that the proposed method not only reflects the simplicity and high speed of the LSSVM but also ensures the accuracy of the prediction intervals by optimizing the model parameters.

    • Chinese Relation Extraction Based on Multi-Granularity and Semantic Information

      2021, 30(3):190-195. DOI: 10.15888/j.cnki.csa.007810

      Abstract (589) HTML (853) PDF 1.03 M (1295) Comment (0) Favorites

      Abstract:Chinese relation extraction adopts character-based or word-based neural networks. Most of the existing methods have word segmentation errors and ambiguity, which will inevitably introduce a lot of redundancy and noise and thus affect the results of relation extraction. In order to solve this problem, this study proposes a Chinese relationship extraction model based on multi-granularity combined with semantic information. In this model, we merge word-level information into character-level information, so as to avoid errors in sentence segmentation; use external semantic information to model polysemous words to reduce the ambiguity caused by semantic words; and adopt Dual attention mechanism at character level and sentence level. The experimental results show that the model proposed in this study can effectively increase the accuracy and recall rate of Chinese relation extraction and has better superiority and interpretability than other baseline models.

    • Automatic Evaluation for English Translation of Literary Works Based on Machine Learning

      2021, 30(3):196-201. DOI: 10.15888/j.cnki.csa.007823

      Abstract (596) HTML (715) PDF 1.25 M (1051) Comment (0) Favorites

      Abstract:The intelligent algorithm model based on machine learning has become the most effective method at present to improve the automatic evaluation for the English translation of literary works. First, the translation rules and particularity of literary works are studied, and the index system of translation evaluation based on the variable features is established. Then, with the aid of the Python language platform, after the English translation is filtered and preprocessed by tools such as Stanford Parser and NLTK, the feature codes and feature degree are obtained with the Vector Space Model (VSM). Furthermore, the results are input into the Random-RF, Original-RF, and AHP-RF algorithm models for training and learning. Thus, the evaluation and analysis of translation quality are completed. The experimental results show that the AHP-RF model combining the analytic hierarchy process, the grey correlation method, and the random forest algorithm has better classification than the other two. Meanwhile, compared with the other four machine translation versions, the manual translation has a high quality score and a low classification error, and the corresponding evaluation results are consistent with the actual translation.

    • Safety Belt Detection Algorithm for Aerial Work Based on Mask R-CNN

      2021, 30(3):202-207. DOI: 10.15888/j.cnki.csa.007812

      Abstract (763) HTML (2292) PDF 1.02 M (2296) Comment (0) Favorites

      Abstract:With the development of computer vision in recent years, more and more attention is paid to the practical application of artificial intelligence algorithms in power security systems. In this paper, aiming at the safety belt specification of power maintenance workers, based on the Mask R-CNN algorithm, we propose a new detection algorithm of safety belts hanging lower than the operator position during aerial work, which can complete the detection of safety belt violation in real-time and efficiently. Furthermore, we propose a new detection method of safety belt violation for aerial work, i.e., Mask-Keypoints R-CNN, which is applicable to the combination of safety belt detection and human key point information. The algorithm cuts the useful safety belt data set from the key parts of human bodies based on the positioning and detection module of the key points of human bodies and judges the violation of operators by combining with the safety belt detection module. In conclusion, the proposed algorithm has strong practicability and high efficiency and has achieved high accuracy.

    • Air Quality Prediction Based on GCN-LSTM

      2021, 30(3):208-213. DOI: 10.15888/j.cnki.csa.007815

      Abstract (1121) HTML (4700) PDF 1.11 M (2689) Comment (0) Favorites

      Abstract:With the development of environmental monitoring technology in China, the grid monitoring system of ambient air quality has received more attention from environmental workers. In order to solve the air quality prediction of small and miniature monitoring stations in the grid monitoring system of air pollution, we propose an air quality prediction model based on GCN and LSTM. First, GCN is applied to extract the spatial features between the small and miniature monitoring stations in the grid monitoring system. Then, LSTM is employed to extract the relevant temporal features. Finally, the linear regression layer is used to integrate the spatial and temporal features and get the prediction results of air quality. Furthermore, experiments are carried out on the air quality monitoring data from 14 small and miniature monitoring stations in Hunnan District, Shenyang, verifying the prediction effect of the proposed model. The experimental results show that the air quality prediction model based on GCN-LSTM is more accurate than the LSTM prediction model in terms of the prediction results on the small and miniature monitoring stations in the grid monitoring with strong spatial association.

    • Trusted Routing Model Based on Opportunistic Networks in VANET

      2021, 30(3):214-220. DOI: 10.15888/j.cnki.csa.007851

      Abstract (550) HTML (905) PDF 1.10 M (1173) Comment (0) Favorites

      Abstract:Because the nodes in VANET are mostly fast-moving vehicles, the mobility of the nodes makes the VANET topology more complicated, the distribution range of the nodes wider, and the potential threat of malicious nodes to the routing gradually greater. These uncertain factors have affected the communication security and space trust values of the vehicle-mounted nodes. In this study, we mainly build a trust evaluation model based on the trust degree of feedback nodes and combine it with the classical opportunistic routing model to propose a trusted routing model better than the existing ones and more suitable for the current complex environment of the VANET. Thus, the security and accuracy of communication between the nodes are further improved. The simulation results show that the performance of each routing model differs significantly under different preset values. Specifically, the FB-SF model increases the detection ratio of malicious nodes as much as possible while improving the accuracy of data transmission.

    • Fault Diagnosis of WSN Nodes Based on RS-CSA-ELM

      2021, 30(3):221-226. DOI: 10.15888/j.cnki.csa.007850

      Abstract (521) HTML (660) PDF 1.21 M (1077) Comment (0) Favorites

      Abstract:In order to discover the hidden troubles of WSN nodes in time and accurately know the running status of WSN, this paper uses the attribute reduction algorithm of rough set theory (RS for short) to reduce the fault attributes of WSN nodes, and reconstructs the training sample data set with the optimal fault attribute decision table as an input to the Extreme Learning Machine (ELM) neural network. In this way, a data-driven fault diagnosis model of WSN nodes is established. The input weights and hidden layer thresholds of the ELM neural network are optimized through Crow Search Algorithm (CSA) to alleviate the unstable output and improve the low classification accuracy of the ELM model caused by the random generation of network parameters. Simulation analysis of the RS-GA-ELM model is carried out. The results show that the RS-GA-ELM model can keep efficiently diagnose faults in data sets with different reliability, which meets the needs of fault diagnosis of WSN nodes.

    • Pedestrian Re-Identification Based on HPLF

      2021, 30(3):227-233. DOI: 10.15888/j.cnki.csa.007841

      Abstract (637) HTML (668) PDF 983.52 K (1220) Comment (0) Favorites

      Abstract:Pedestrian re-identification generally considered as a sub-problem of image retrieval. Due to the distance between the camera and the pedestrian, the definition of the pedestrian photo is generally fuzzy, and the camera’s view angle of pedestrians is fixed, so it is not enough to recognize pedestrians by faces. In order to better mine strong local features and improve the accuracy of pedestrian re-identification, this study proposes an algorithm, namely Horizontal Pooling for Local Feature (HPLF). We preprocess the input joint data set in ResNet-50 network, extract features, and horizontally cut the feature map generated by ResNet-50 network, with which we calculate the distance between every two features. Triple loss with hard example mining (TriHard loss) is used for training as a local feature loss function. The global distance is calculated according to the feature map and trained through TriHard loss. The two loss functions plus a Softmax cross entropy loss function are combined as the total loss function for parameter correction. The experimental results show that HPLF’s performances of mean Average Precision (mAP), Rank-1, Rank-5, and Rank-10 in the Market1501 data set are about 3% higher than those of other algorithms.

    • Miao Pattern Design Based on Extensional Representation and Neural Network

      2021, 30(3):234-242. DOI: 10.15888/j.cnki.csa.007829

      Abstract (527) HTML (588) PDF 2.51 M (1051) Comment (0) Favorites

      Abstract:With regard to the cultural inheritance and design application of Miao patterns, an innovative design method of ethnic patterns based on extensional representation and neural network is proposed to deconstruct, map, and reconstruct Miao batik patterns. Firstly, Miao batik patterns are characterized by extension, and the divergence tree is used to construct the design growth stage model to expand and analyze the elements of Miao patterns. Secondly, Miao batik patterns are analyzed with Kansei images based on Kansei engineering, proposing a pattern deconstruction method for pattern configuration, pattern semantics and types, with which the deconstructive space of pattern feature elements and the cognitive space of emotional images are constructed. The neural network is used to construct the Kansei prediction model that recommends design elements such as pattern configuration to users in terms of their preference, and the design thinking is converging. Its advantages are verified by the comparison with the linear regression prediction model. Finally, shape grammar is used to refine Miao batik patterns according to the characteristic elements recommended by the model. The method is verified feasible with Miao batik patterns and can provide a reference for the deconstruction and innovative design of other ethnic patterns.

    • Prediction on Keywords Popularity of Public Opinion Based on Factor Analysis and Elman Network

      2021, 30(3):243-249. DOI: 10.15888/j.cnki.csa.007811

      Abstract (538) HTML (710) PDF 2.87 M (1110) Comment (0) Favorites

      Abstract:It is helpful for institutions to master the whole trends of target group that research on keywords popularity of network public opinions from a macroscopic perspective, precisely formulating corresponding strategies to enhance the level of opinion guidance. With Sina Weibo data set as an example, Factor Analysis (FA) is used to mine the internal factors of public opinions; a model that analyzes and predicts the keyword popularity of network public opinions is created through the initial parameters optimized by Genetic Algorithm (GA) and Elman network structure. The results show that predictions made by our method is more precise than those of original data sets and standard Elman network. Thus, it can be applied to providing reference for decision making.

    • Air Quality Index Prediction Based on Outlier Detection and Error Correction

      2021, 30(3):250-255. DOI: 10.15888/j.cnki.csa.007817

      Abstract (708) HTML (1017) PDF 1.05 M (1310) Comment (0) Favorites

      Abstract:Air Quality Index (AQI) prediction can provide guidance for people’s daily production activities and air pollution control. In view of the problem that AQI prediction model is greatly affected by outliers, the isolation forest algorithm is used to detect outliers in the air quality data set; the Outlier Robust Extreme Learning Machine (ORELM) model is proposed for AQI prediction, and an error correction module is constructed to correct model prediction error. Finally, with the air quality data of Beijing as the research object for empirical analysis, the ORELM model and the Extreme Learning Machine (ELM) model are used to make predictions, and the prediction error of the ORELM model is corrected. Experimental results show that the ORELM has stronger generalization performance for outlier data sets, and the error correction module can effectively improve the prediction accuracy of the model.

    • Division of Metro Operation Periods Based on Feature Clustering of Passenger Flow

      2021, 30(3):256-261. DOI: 10.15888/j.cnki.csa.007799

      Abstract (676) HTML (1002) PDF 1012.76 K (1492) Comment (0) Favorites

      Abstract:An accurate and reasonable division scheme of operation periods is the premise and foundation for formulating metro train operation plans, and it is also an important way to improve metro operation efficiency. The feature vectors of passenger flow are constructed to divide metro operation periods. At an interval of 10 min, the operation period in a day is segmented, and the feature vectors of all the periods are constructed according to the characteristics of passenger flow changes in their corresponding periods. The K-means algorithm is used for clustering, and the results are evaluated by cluster evaluation indicators such as elbow method and silhouette coefficient to determine the optimal number of clusters, obtaining the optimal division scheme of operational periods. Finally, the division scheme of operation periods of Line 1 of Fuzhou Metro is given as an example, which verifies the feasibility of this method.

    • Application of Auto-Encoder in Optimization of Water Quality Monitoring Points

      2021, 30(3):262-266. DOI: 10.15888/j.cnki.csa.007833

      Abstract (590) HTML (554) PDF 1.64 M (996) Comment (0) Favorites

      Abstract:Water is one of the necessary elements for our human survival, and the results of water quality monitoring are the basis for water quality control. In a region or watershed, many water quality monitoring points can be found in a region or watershed. With population growth, industrial development, and soil variety, water environment has undergone drastic changes, and some points may be wrongly, overly, or repetitively selected. As for this, resource-saving measures need to be taken to comprehensively show the distribution of water quality with as few points as possible. In this study, a method that combines auto-encoder neural network with hierarchical clustering is proposed. This method uses auto-encoder for feature selection of input samples and analyzes the samples after feature dimensionality reduction through hierarchical clustering, optimizing water quality monitoring points. The experiment results show that the method is more effective as opposed to the method of fuzzy clustering without feature dimensionality reduction.

    • Performance Evaluation and Type Selection of Dredging Pump Based on Entropy Weight-Fuzzy Analytic Hierarchy Process

      2021, 30(3):267-271. DOI: 10.15888/j.cnki.csa.007838

      Abstract (636) HTML (584) PDF 787.63 K (995) Comment (0) Favorites

      Abstract:In the process of removing the jacket platform with the internal cutting method, the pile leg needs to be dredged. The dredging pump is the key equipment of the dredging system in the pile leg, and its performance directly determines the work efficiency of the dredging system. The performance of dredging pumps is comprehensively evaluated based on entropy weight–fuzzy analytic hierarchy process to obtain the optimal configuration scheme of dredging pumps in the dredging system. A mathematical model for evaluating the performance index of dredging pumps was established, and the most preferred scheme was obtained from five dredging pumps with similar performance. The research results show that the NSQ100-60-45 dredging pump has the largest comprehensive evaluation weight, and its performance is more suitable for mud pumping operations in the legs. The performance evaluation model in this study is highly reliable and can give guidance to the type selection of dredging pumps in the dredging system.

    • Multi-Modal Spatiotemporal Action Recognition Based on Deep Learning

      2021, 30(3):272-275. DOI: 10.15888/j.cnki.csa.007840

      Abstract (688) HTML (1048) PDF 591.10 K (1331) Comment (0) Favorites

      Abstract:In view of the time-series difficulty in video understanding and a large amount of calculation in traditional methods, we propose a method with spatio-temporal module for action recognition. With a residual network as the framework, this method adds spatio-temporal module to extract images and time series, adds RGB difference to enhance data, and finally uses the NetVLAD method to aggregate all feature information. In this way, actions are classified. The experimental results show that the multimodal method based on spatio-temporal module has better recognition accuracy.

Current Issue


Volume , No.

Table of Contents

Archive

Volume

Issue

联系方式
  • 《计算机系统应用》
  • 1992年创刊
  • 主办单位:中国科学院软件研究所
  • 邮编:100190
  • 电话:010-62661041
  • 电子邮箱:csa (a) iscas.ac.cn
  • 网址:http://www.c-s-a.org.cn
  • 刊号:ISSN 1003-3254
  • CN 11-2854/TP
  • 国内定价:50元
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.

Beijing Public Network Security No. 11040202500063