• Volume 28,Issue 10,2019 Table of Contents
    Select All
    Display Type: |
    • End-to-End Speech Separation Based on Deep Acoustic Feature

      2019, 28(10):1-7. DOI: 10.15888/j.cnki.csa.007093

      Abstract (1274) HTML (1626) PDF 1.17 M (1960) Comment (0) Favorites

      Abstract:An end-to-end single channel speech separation algorithm based on deep acoustic feature is proposed. The traditional acoustic feature extraction methods require the Fourier transform, discrete cosine transform and other operations. This will cause speech energy loss and long latency. In order to improve these problems, the original waveform of the speech signal is used as an input to a deep neural network, deeper acoustic features of the speech signal are learned through a network model. Objective evaluation shows that the proposed algorithm not only improves the performance of speech separation effectively, but also reduces the time delay of speech separation algorithm.

    • Research on Semi-Automatic Generation Technology of Object Detection Datasets

      2019, 28(10):8-14. DOI: 10.15888/j.cnki.csa.007101

      Abstract (1184) HTML (858) PDF 1.19 M (1602) Comment (0) Favorites

      Abstract:Object detection is widely used in the field of computer vision. In different occasions, we need to use different training set to train the model. However, manually generating label is very time consuming. This study proposed a semi-automatic method to generate labels for dataset, then automatically filter them according to the threshold set by image similarity, lastly retain the required images and corresponding labels as the final dataset. Experiments show that the method can both improve the speed and ensure accuracy rate of generating labels for dataset.

    • Pulse Wave Recognition Using Deep Hybrid Neural Networks Based on GoogLeNet and ResNet

      2019, 28(10):15-26. DOI: 10.15888/j.cnki.csa.007110

      Abstract (1465) HTML (966) PDF 2.08 M (2577) Comment (0) Favorites

      Abstract:To improve the accuracy of pulse wave recognition, MIRNet2 is proposed, which is a kind of modified deep hybrid neural networks. Firstly, processable data sets of Caffe are obtained by main pulse extraction, segmenting cycle and making hdf5 data sets. Secondly, deep hybrid neural networks are designed. Inception-ResNet (IRNet) is consisted of inception modules and residual modules, containing IRNet1, IRNet2 and IRNet3. Subsequently, Modified Inception-ResNet (MIRNet) composed of modified Inception modules, residual modules and pooling modules (or reduction modules) is proposed, including MIRNet1 and MIRNet2. Compared with other neural networks in the study, MIRNet2 is the best one, with the specificity of 87.85%, the sensitivity of 88.05% and the accuracy of 87.84%, respectively. In addition, parameters and operations of MIRNet2 are also less than that of IRNet3.

    • Satellite Image Recognition and Classification Method Based on Deep Learning

      2019, 28(10):27-34. DOI: 10.15888/j.cnki.csa.007081

      Abstract (1473) HTML (3316) PDF 1.57 M (2025) Comment (0) Favorites

      Abstract:Satellite remote sensing technology is a very important geo-spatial monitoring technology. After being processed, the satellite remote sensing images have a large amount of data characteristics of various complex data types, the traditional target classification and recognition ways spend a lot of manpower and material resources. In order to reduce the workload and provide convenience for subsequent processing, we consider using deep learning algorithms for satellite images classification and recognition. In this paper, we designed an image recognition and classification method based on VGGNet. We augmentated data by using haze removal algorithm and other tricks. And we added ridge regression to use correlations between labels to predict. Verified by experiment comparison, this method can achieve more than 90% of F2 score. Finally, an online recognition, classification and display system based on Django is built by using this method.

    • Vulnerability Detection of Device Drivers Based on Pair Functions’ Calling Context

      2019, 28(10):35-44. DOI: 10.15888/j.cnki.csa.007099

      Abstract (979) HTML (594) PDF 1.43 M (1312) Comment (0) Favorites

      Abstract:Since the device drivers of Linux work in the kernel mode, in this specific work scenario, the vulnerability caused by the device drivers can easily affect the stability and security of the operating system. At present, the most proportion of various types of device drivers' vulnerabilities is resource operation vulnerability. In this case, a vulnerability device detection method of device drivers based on pair functions' calling context is proposed. Firstly, we introduced the concept of pair function, according to which the automatic extraction and optimization of the pair function were performed for the specific drivers. Then the execution path of the pair function in the resource request and release process was recorded based on manual analysis results. Finally, the pair function was combined with the corresponding calling context scenario to verify whether the application and release of memory resources in the device driver matched in the hierarchy exactly. In order to verify the effectiveness of this method, vulnerability detection method was applied to different drivers in the experiment, and the corresponding false negative, false positive, and coverage were recorded. The experimental results show that the device drivers' vulnerability detection method has higher accuracy and faster detection speed, and the method does not depend on conditions such as real-time compilation and hardware devices.

    • Recommendation System Based on Real-Time Recommend and Offline Recommend

      2019, 28(10):45-52. DOI: 10.15888/j.cnki.csa.007087

      Abstract (1414) HTML (4103) PDF 1.17 M (1745) Comment (0) Favorites

      Abstract:Recommendation system is a tool to automatically find information that users may be interested in from a large amount of information. How to get closer to users' preferences, satisfy users' long-term inherent preferences, and simultaneously take into account users' short-term interest focus changes is an everlasting research problem of recommendation systems. In addition, in order to improve the recommended performance when designing the system, we not only focus on user modeling optimization, recommendation object modeling optimization or recommendation algorithm optimization, but also need to systematically study the recommendation system as a whole, focusing on system fluency and scalability. To solve these problems, this study designs a recommendation system that combines real-time recommendation with offline recommend, and proposes a method to ensure the fluency of the system by using the pool of recommendation data. Based on the analysis of real-time data and historical data, real-time recommendation and offline recommendation are provided, which can fit the long-term preferences of users and adapt to the recent change of interest focus. The control module of the system is used to control and adjust different recommendation result data to improve the scalability of the system. Based on the recommendation system, this study conducts a recommendation experiment for WeChat articles, and evaluates the recommendation effect by analyzing the data in the recommendation pool. The experimental results show that the recommendation data of the system can be gradually close to the user's interest preference.

    • Generating Auxiliary Diagnosis Dialogue System for Gastroenterology

      2019, 28(10):53-60. DOI: 10.15888/j.cnki.csa.007128

      Abstract (1004) HTML (592) PDF 1.23 M (1467) Comment (0) Favorites

      Abstract:The rapid development of society brings more and more pressure to people. Due to work pressure and self-problems, more and more people are eating three meals irregularly and unhealthily, which leads to the growing population of people suffering from digestive diseases. When the body just appears abnormal, most people first will choose to find information on the Internet. Due to the limitations of traditional search engines, the process is time consuming, and because of the diversity of diseases, it is difficult for users to accurately obtain relevant information. In view of this problem, considering that the dialogue system is a more advanced information retrieval system. This study explores a generative dialogue system suitable for the field of gastroentology, using support vector machine and active learning to obtain professional consultation dialogue corpus of gastroenterology on multiple medical websites. Labor and statistics are combined to build a professional dictionary of diseases, drugs, and symptoms of digestive diseases, word segmentation is improved in the medical field. On this basis, multi-Encoder and multi-Decoder structure is combined with gated loop unit GRU, the attention mechanism is also added, then the model strengthening training method is proposed, which combines reverse input and key-value pair vector and Word2Vec vector, to obtain the final model. The experimental results show that the word segmentation result is much higher than the traditional method, and the resulting dialogue model can effectively generate the sentence related to the question, which can improve the answer accuracy of the dialogue system.

    • Secured Data Self-Destructive Service Platform Based on WeChat

      2019, 28(10):61-67. DOI: 10.15888/j.cnki.csa.007131

      Abstract (1355) HTML (832) PDF 1.18 M (1706) Comment (0) Favorites

      Abstract:Facing precarious status quo of that users' information data in social software is easily leaked, we have carried out a research on secured self-destructive service platform of social software for data. We make full advantage of WeChat, combining TLS protocol, a self-destructive method based on network and OAuth2.0, an open protocol to allow secure authorization from the third party webpage of WeChat to construct a secured self-destructive service platform based on Wechat. We ensure the safety of data transmission between client and server by TLS protocol, take a modified method based on network as the secured self-destructive scheme for data on the server side and make the advantage of OAuth2.0 which is a secure authorization from the third party webpage of WeChat to accomplish users' identification. theories and example analysis verifies the feasibility and safety of this design plan

    • Building Energy Consumption System Based on B/S Architecture

      2019, 28(10):68-73. DOI: 10.15888/j.cnki.csa.007092

      Abstract (1271) HTML (643) PDF 1.08 M (1504) Comment (0) Favorites

      Abstract:With the development of China's economy, the area of office buildings and large-scale public buildings is increasingly growing, resulting in ever-increasing high energy consumption, such as electricity consumption, water consumption, and gas consumption. This system based on B/S structure and Webx frame technology can conduct statistical analysis of energy consumption. The collection program uses Go language technology to accquire data through network communication and serial device server which is then connected to the terminal device through the RS485 serial interface, and collects data by protocols such as modbus or opc. In addtion, data storage is deployed using MySQL cluster. Such modern information technologies help us to know the current energy consumption of the office building in real time. Refering the previous system design scheme, we then further upgrades and rebuilds technical selection, big data storage, acquisition system modularization, hardware performance, and other aspects to make the system easier to be maintained and extended, and meanwhile adds distributed deployment and so on in attempting to improve the load capacity. In terms of technical selection, the most popular and stable architecture scheme is selected. Moreover, an open-source and free MySQL cluster mode is used to replace the expensive Oracle database for database storage.

    • Real-Time Analysis Storage System for Fake Plate Vehicle Based on Kafka and Storm

      2019, 28(10):74-79. DOI: 10.15888/j.cnki.csa.007094

      Abstract (1098) HTML (974) PDF 1.14 M (1208) Comment (0) Favorites

      Abstract:With the increase of the number of motor vehicles and vehicle traffic in the city, the phenomenon of fake plate vehicle appears repeatedly. In order to solve the problem of fake plate monitoring, the traffic management departments adopt traditional identification methods, such as manual identification, license plate recognition, radio frequency identification, etc. Nevertheless, facing the massive log records, these methods generally have problems of low efficiency and poor real-time performance. So big data technology was introduced, and fake plate vehicle real-time analysis storage system based on Kafka and Storm was proposed. Kafka can be used as a middleware for caching, improving the synchronization of data collection and data analysis, and avoiding data loss. The Storm framework can realize real-time calculation of log information, and then store the information of the fake plate vehicles in the specified document. The entire system has real-time, distributed storage, stability, scalability, and so on.

    • Design and Implementation of E-Commerce Safety Module Based on Added Salt BCrypt Algorithm

      2019, 28(10):80-85. DOI: 10.15888/j.cnki.csa.007108

      Abstract (1067) HTML (848) PDF 1.09 M (1681) Comment (0) Favorites

      Abstract:With the advent of the era of big data, the situation of password leakage occurs, data security has become an increasingly concerned issue. Using springmvc+spring+mybatis framework technology, this paper expounds in detail the application of added salt BCrypt algorithm in e-commerce security module by means of model analysis, database table design, and timing diagram logic jump, which effectively solves the drawbacks of MD5 encryption algorithm, and greatly improves the security of information.

    • Improving Hybrid Location Recommendation System Based on Spark Parallelization

      2019, 28(10):86-91. DOI: 10.15888/j.cnki.csa.007118

      Abstract (1110) HTML (534) PDF 1.04 M (1309) Comment (0) Favorites

      Abstract:The recommendation algorithm is one of the most important algorithms in data mining. Location recommendation is an important research content of the recommendation system. Aiming at the problems of sparse data, cold start and low degree of personalization, the improved hybrid location recommendation algorithm based on Spark parallelization is designed and implemented. The algorithm combines content-based recommendations and collaborative filtering-based recommendations, combines the user's current preferences with the opinions of other users. We improve data sparsity by using a matrix fill based on user preferences for location attributes; Also, for massive data, the system uses Spark distributed cluster to realize parallel computing, which shortens the model training time. Experimental results show that compared with other recommended algorithms, the proposed algorithm can effectively improve data sparsity and improve recommendation.

    • Liver Deformation Simulation System in Virtual Surgery

      2019, 28(10):92-97. DOI: 10.15888/j.cnki.csa.007105

      Abstract (1364) HTML (1326) PDF 965.94 K (1502) Comment (0) Favorites

      Abstract:Liver has complex biomechanical properties. Therefore, it is difficult to achieve real-time requirement because of the large amount of calculation in deformation simulation. At the same time, it is inaccessible to achieve the reality in real-time simulation of liver deformation. In order to solve this contradiction, a hybrid model which can automatically change the operating area and force of surgical instruments is developed. According to theoretical analysis, a construction method for the mesh and meshless model of liver is proposed. The experimental results show that the algorithm has high computational efficiency and good deformation effect, which can meet the realistic and real-time requirements of virtual liver surgery simulation. The conclusions drawn in this paper can be used to guide the study of virtual liver soft tissue surgery.

    • Highly Trusted Interaction Framework for Business Behavior Control Based on Network Isolation Architecture

      2019, 28(10):98-102. DOI: 10.15888/j.cnki.csa.007083

      Abstract (1239) HTML (567) PDF 875.15 K (1395) Comment (0) Favorites

      Abstract:This study proposes a highly trusted interaction framework for business behavior control based on network isolation architecture. This framework not only makes the access of business data in the high security zone under the complex security architecture for enterprise internet mobile applications possible but also ensures the security of key data of the business system. Under the requirements of network security protection, the mobile access gateway is introduced to decompose the interaction process of business data across the security zone. Then, it designs an access conversion and communication method, which realizes the safe and reliable transmission of business data through various isolation devices and business flow control. The framework has been widely used in many business areas such as employee reimbursement, attendance punching, power system distribution repair and mobile inspection.

    • Shell Model Construction and Optimization of 3D Printing Based on Stress Distribution

      2019, 28(10):103-111. DOI: 10.15888/j.cnki.csa.007120

      Abstract (1216) HTML (570) PDF 1.71 M (1826) Comment (0) Favorites

      Abstract:In recent years, the mature 3D printing technology has brought the distance between model design and product manufacturing closer. However, the high material cost is still an important factor restricting its development. Therefore, how to optimize the model structure without changing the appearance of the model, so as to reduce the printing volume of the model and cut down the printing cost is an urgent problem to be solved. For this problem, we propose a shell model construction and optimization algorithm based on stress distribution. This algorithm first constructs a distance field based on the voxel representation of the model and extracts the initial uniform thickness shell model. Then, this algorithm expands the inner surface outward adaptively, based on the von Mises stress value. The algorithm stops when the preset constraints are reached. The final optimization model is enclosed by the new inner surface and the original outer surface. Experimental results show that, the optimized shell model volume is 17.2%~24.4% of the input model volume while satisfying the constraints of appearance and mechanical stability, which reduces the printing volume greatly and lower the printing cost effectively.

    • Two-Stage Data Fusion Model and Algorithm Based on Environmental Monitoring

      2019, 28(10):112-119. DOI: 10.15888/j.cnki.csa.007073

      Abstract (1320) HTML (1084) PDF 1.64 M (1445) Comment (0) Favorites

      Abstract:The data collected by multi-source sensors not only have a lot of redundancy, but also affect the final monitoring results. In order to improve the accuracy of monitoring, this study proposes a two-level data fusion model and algorithm for grassland environment monitoring. In the first-level data fusion, the adaptive weighted averaging method is used to fuse the similar sensors in each region, and then the BP neural network is used to train and fuse the heterogeneous sensors in the region, thus a preliminary judgment on the environmental conditions of each region is obtained. Because of the uncertainty of the fusion result by BP neural network, the secondary fusion uses DES evidence theory to analyze the primary fusion result and get the decision-making judgment of grassland environment. Finally, the validity and analysis of the model and algorithm are carried out. The experimental results show that the proposed method can accurately monitor the grassland environment. At the same time, it provides some valuable guidance and decision-making basis for the efficient management and scientific conservation of grassland environment.

    • Image Stitching Algorithm Based on Structural Information

      2019, 28(10):120-129. DOI: 10.15888/j.cnki.csa.007098

      Abstract (1267) HTML (741) PDF 1.79 M (1469) Comment (0) Favorites

      Abstract:Two or more image stitching problems with more texture and higher noise were studied. When two images are spliced, the extraction of image feature points has a great influence on the image splicing results. The classical SIFT algorithm is a better local feature point extraction algorithm. For images with more texture and higher noise, the SIFT algorithm extracts a large number of feature points, which affects the accuracy and speed of matching. In this study, based on structural information-based image matching algorithm called SKM, the RTV algorithm is used to extract the information structure of the image, and the texture noise in the image is effectively removed. After denoising, the SIFT algorithm is used to extract feature points for matching. Finally, RANSAC algorithm is used to screen matching points to improve the accuracy. The transformation matrix H obtained by SKM algorithm acts on the original image to complete image mosaic.

    • Facial Image Retouching Detection Algorithm Based on Multi Scale-Convolutional Neural Network

      2019, 28(10):130-137. DOI: 10.15888/j.cnki.csa.007072

      Abstract (1455) HTML (732) PDF 1.36 M (1509) Comment (0) Favorites

      Abstract:In order to solve the problem that the existing facial retouching detection algorithm has complex feature extraction and low recognition rate, we proposed a facial retouching detection algorithm based on Multi-Scale-Convolutional Neural Network (MS-CNN). Different from the traditional CNN, MS-CNN adds image preprocessing, which uses Histograms of Oriented Gradient (HOG) feature-based facial extraction algorithm to extract the facial part from the original image. It connects the Local Response Normalization (LRN) layer after the first pooling layer to accelerate the convergence of the model. A multi-scale convolution layer is proposed, which cascades convolution kernel of 1×1, 3×3, and 5×5 to improve the classification accuracy. The experimental results show that the detection accuracy of the proposed algorithm is 99.5% in LFW data set and 92.9% in ND-ⅢTD data set, respectively. Compared with the mainstream network structure and existing facial retouching detection algorithms, the detection accuracy of the proposed algorithm is significantly improved.

    • Method and Improvement of Maximum Search Based on PRAM Parallel Model

      2019, 28(10):138-144. DOI: 10.15888/j.cnki.csa.007119

      Abstract (1118) HTML (599) PDF 1.26 M (1497) Comment (0) Favorites

      Abstract:With the emergence of multiprocessors, parallel technology has attracted widespread attention and become an important technology to speed up the processing of problems. Nevertheless, the use of parallel technology to speed up computing has also led to a sharp increase in the number of processors and a significant increase in parallel costs. To solve this problem, by studying the shortcomings of three parallel maximum search algorithms based on PRAM (Parallel Random Access Machine), a parallel maximum search algorithm based on data partitioning method is proposed, which is better than the balanced tree algorithm, fast search method and double logarithmic depth tree method. The maximum searching algorithm based on data partitioning method effectively solves the problems of uneven workload allocation, excessive demand for processors and harsh implementation conditions in existing parallel methods. This provides a direction for similar parallel algorithms to reduce parallel costs.

    • Improved Classification Algorithm Based on Support Vector Machine

      2019, 28(10):145-151. DOI: 10.15888/j.cnki.csa.007080

      Abstract (1814) HTML (1802) PDF 1.24 M (1461) Comment (0) Favorites

      Abstract:In order to improve the accuracy and generalization ability of Support Vector Machine (SVM) classification, this paper presents an improved binary tree classification algorithm based on SVM. It introduces basic principle of SVM, and summarizes multi-classifier classification algorithm and characteristics. Combining the advantages of the classification algorithms and introducing different weights for the classifier, this study proposes improved classification algorithm of the binary tree, which effectively avoids the shortage of common classification algorithms. Simulation experiments and comparison of the typical multi-class classification algorithms verify that the algorithm is effective. The algorithm provides an effective way for multi-class classification prediction research.

    • Short-Term Traffic Flow Forecasting Model Based on LSTM-BP

      2019, 28(10):152-156. DOI: 10.15888/j.cnki.csa.007077

      Abstract (2237) HTML (814) PDF 1.32 M (2582) Comment (0) Favorites

      Abstract:In order to alleviate the increasingly serious traffic congestion problem, realize intelligent traffic control, provide accurate real-time traffic flow prediction data for traffic flow induction and traffic travel, an LSTM-BP combined model algorithm based on long-short-time memory neural network (LSTM) and BP neural network is designed. Mining the characteristic factors of known traffic flow data, establishing the framework of time series prediction model, and using Matlab to complete the simulation from the data processing to the model simulation to realize the accurate prediction of short-term traffic flow based on LSTM-BP. Compared with the three prediction network models of LSTM\BP\WNN, the results show that the time series predicted by LSTM-BP has higher accuracy and stability. The construction of the model can provide basis and reference for the prediction of traffic distribution, the division of traffic modes, and the distribution of real-time traffic flow.

    • Surface Reconstruction Algorithm of Medical Data Based on Improved Total Convolution Neural Network

      2019, 28(10):157-163. DOI: 10.15888/j.cnki.csa.007095

      Abstract (966) HTML (708) PDF 1.25 M (1294) Comment (0) Favorites

      Abstract:In order to realize the rapid detection and classification of medical data, the surface reconstruction design of medical data needs to be carried out. A surface reconstruction algorithm of medical data based on improved total convolution neural network is proposed. The big data sampling of medical data was carried out by using radio frequency identification technology, the medical data collected by RFID was processed by information fusion, and the correlation statistical characteristics of medical data were extracted by multiple regression analysis. According to the characteristics of medical data, matching filter detector is used for redundant filtering, and phase space reconstruction technology is used to reconstruct medical data after purification. In order to realize the surface reconstruction and automatic recognition of medical data, an improved total convolution neural network classifier is used to classify and recognize the reconstructed data. The simulation results show that the proposed method has a good effect on the redundant feature processing of medical data. The accuracy of data classification is more than 90%, and the error of medical data reconstruction is smaller and the time consuming is less.

    • Rib Suppression Algorithm Based on U-Net in Chest Radiographs

      2019, 28(10):164-169. DOI: 10.15888/j.cnki.csa.007100

      Abstract (1535) HTML (797) PDF 1.21 M (1761) Comment (0) Favorites

      Abstract:Skeletal structure and human organs overlap in the chest X-ray films, which has a negative impact on the intelligent detection system of doctors and pulmonary nodules, because X-ray image only has two-dimensional information. Restraining rib structure in the image can improve the above situation to a certain extent. We regard ribs as noise information in images, and use image denoising method to suppress ribs. In this study, we use deep convolution network as the basic model, and try to improve the performance of the model by analyzing and trying a variety of strategies. Ultimately, we use U-net network structure to enhance the performance of network details through jump connections and residual learning strategies. Experiments prove that the proposed method can effectively suppress the adverse effects of rib structure in X-ray images, and improve the performance of pulmonary nodule detection tasks.

    • Medical Image Fusion Algorithm Based on Non-Subsampled Shearlet Transform and Feature Synthesis

      2019, 28(10):170-177. DOI: 10.15888/j.cnki.csa.007109

      Abstract (1331) HTML (990) PDF 2.18 M (1540) Comment (0) Favorites

      Abstract:Aiming at the problem that the detailed texture is not clear enough for the fused medical image, this study proposes a new medical image fusion algorithm on the basis of non-subsampled shearlet transform (NSST) to fuse the multimodal medical image to enhance the detail structure extraction, improve fused image quality and provide a basis for medical diagnosis. First of all, the registered source image is decomposed by NSST to obtain a low-frequency sub-band and a series of high-frequency sub-band. Then, for the low-frequency sub-band coefficients, this study proposes a fusion method using sub-band selection between the regional average energy and regional standard deviation. For high-frequency sub-band coefficients, the fusion method is performed using the new sum of modified Laplacian (NSML). Afterwards, the fused low-frequency, high-frequency sub-band coefficients are inversely transformed by NSST to obtain a fused image. Finally, a large number of experiments were performed on grayscale and color medical multimodal images, and IE, SF, SD, and AG were selected to evaluate the fused images. The simulation results show that the proposed algorithmimprove subjective visual effect and objective evaluation. Compared with other algorithms, the average values of IE, SD, SF, and AG increased by 2.99%, 4.06%, 1.78% and 1.37%, respectively. The fused image contains more detailed texture information and better visual effect.

    • Data Encryption Transmission Algorithm Based on MQTT

      2019, 28(10):178-182. DOI: 10.15888/j.cnki.csa.007124

      Abstract (2334) HTML (4066) PDF 795.74 K (5863) Comment (0) Favorites

      Abstract:An improved data transmission encryption algorithm MQTT-EA (MQTT Encryption Algorithms) is proposed. In this algorithm, the IoT device and server sides randomly generate their own private keys, then notify each other of their own private keys and combine them into the final session master key through the algorithm. The secure data is transmitted through DES encryption and decryption. The data transmission process is simulated by rivals A and B, and MQTT-EA is proved to be secure as long as the session key generation algorithm is not leaked.

    • Breakout Local Search for Scheduling Freight Trains with Synchronized Transferring Operations

      2019, 28(10):183-189. DOI: 10.15888/j.cnki.csa.007096

      Abstract (1030) HTML (543) PDF 1.01 M (1429) Comment (0) Favorites

      Abstract:Railway container terminals are regarded as the important nodes in hinterland transportation network. Transferring operation is main activity in operations management of container terminals. Scheduling the transferring activities optimally can be able to shorten the operating cycle among different transport vehicles, and improve the efficiency of cargo transportation in rail multimodal transportation. To make the operation plan of stacking yard rapidly, the transferring operation requires the container movement must be stacked intermediately via yard at present. In that scenario, two loading/discharging activities and one stacking activity must be occurred. For avoiding idly extra activities, synchronized transferring should be recommended. In this study, a mixed integer programming model with maximizing the number of synchronized transferring containers is proposed. Due to its intractability, a breakout local search algorithm is designed to solve this problem under study, especially for large-sized instances. The computational experiments are performed to evaluate the performance of the proposed algorithm.

    • Chinese Text Proofreading Method Based on Neural Network and Attention Mechanism

      2019, 28(10):190-195. DOI: 10.15888/j.cnki.csa.007097

      Abstract (1246) HTML (1103) PDF 1.15 M (1620) Comment (0) Favorites

      Abstract:Chinese text proofreading is one of the key tasks in Chinese natural language processing, and manual proofreading is difficult to meet the data volume requirement of daily work, and the text proofreading method based on statistics can not deal with semantic errors flexibly. Aiming at the above problems, a Chinese text proofreading method based on neural network and attention mechanism is proposed. The bidirectional Gated Recurrent Unity neural network layer is used to obtain text information and feature extraction, and the ability of attention mechanism layer to enhance the semantic logic relation between words is introduced. The model is implemented under the framework of deep learning based on Keras. Experimental results show that this method can proofread text with semantic errors.

    • Research on Genetic Algorithm for Scheduling of Parallel Batch Processing Machines with Non-Identical Job Sizes

      2019, 28(10):196-200. DOI: 10.15888/j.cnki.csa.007117

      Abstract (889) HTML (512) PDF 1011.04 K (1292) Comment (0) Favorites

      Abstract:This study considers the application of genetic algorithm for scheduling of parallel batch processing machines with non-identical job sizes. Jobs have different sizes and release times. Firstly, we propose a mathematical programming model based on the hypothesis of the problem, and use BF and ERT-LPT to implement batch scheduling of jobs. Secondly, since the problem considered is NP-Hard, we design a new selection, crossover and mutation operation and solve it with genetic algorithm. Finally, the effectiveness of the algorithm through simulation experiments is verified.

    • Data Augmentation Method Based on Generative Adversarial Network

      2019, 28(10):201-206. DOI: 10.15888/j.cnki.csa.007107

      Abstract (3009) HTML (8053) PDF 1.07 M (3206) Comment (0) Favorites

      Abstract:Deep learning has revolutionized the performance of classification, but meanwhile demands sufficient labeled data for training. Given insufficient data, neural network is apt to overfitting, which is quite general in low data regime. We propose a data augmentation technique based on generative adversarial network to address the network training and data shortage problem. The experimental results show that the synthesized data has semantic similarity compared with the real data, and at the same time it can present the diversity of the context. After adding the synthesized data, the neural network can be trained more stably, and the accuracy of the classification is further improved. Comparing the proposed algorithm with some other data augmentation techniques, the proposed method has the best performance, which proves the feasibility and effectiveness of this technique.

    • Vehicle Logo Recognition Method of Feature Fusion Based on D-S Evidence Theory

      2019, 28(10):207-212. DOI: 10.15888/j.cnki.csa.006757

      Abstract (1050) HTML (1073) PDF 1.08 M (1275) Comment (0) Favorites

      Abstract:For the intelligent traffic, there are inaccuracies in the multi-dimensional information identification of vehicles. Especially for vehicle logo recognition, the recognition results depend largely on high-resolution and high-quality images. A new vehicle logo identification method is proposed for distinguishing low-quality vehicle image captured at the bayonet. This method is based on the feature fusion of D-S evidence theory, extracts Hu invariant moments and HOG features, and uses different classifiers,the basic probability distribution (BPA) is constructed, the improved D-S evidence theory is used to fuse, and the final recognition result is given according to the discriminant rule. Through experiments, it is proved that the accuracy can be maintained at a low resolution, and the classification accuracy is 94.29%, which is more robust than a single feature recognition.

    • Elastic Scaling Strategy Based on Kubernetes Application

      2019, 28(10):213-218. DOI: 10.15888/j.cnki.csa.007106

      Abstract (1063) HTML (1344) PDF 974.34 K (1562) Comment (0) Favorites

      Abstract:Autoscaling is a key feature of cloud computing. It can expand computing resources in time according to application workload and achieve load balancing under high concurrent requests. Container-based micro-services should also have the function of autoscaling so as to have stably performance under different workloads. The elastic scaling algorithm of Kubernetes, a widely used container layout tool, has unsatifactory flexibility. Pod will expand frequently to deal with sudden traffic, and the scaling degree can not meet the current load requirements, which will make a system instability. To solve this problem, an automatic scaling mechanism is proposed, which combines the response expansion with the elastic scaling tolerance, and ensures the reliability of the system. Our method greatly improves the flexibility of the system, and is also competent when facing high application load. Experiments results show that when the system meet with heavy traffic and high concurrent requests, the failure request rate can decrease by 97.83% after carrying out the proposed method. So our method can ensure the stability of the system and realizes the load balancing of the application well.

    • Research on Mimic Data Processing Based on Mycat Middleware

      2019, 28(10):219-225. DOI: 10.15888/j.cnki.csa.007079

      Abstract (1570) HTML (526) PDF 1.49 M (1739) Comment (0) Favorites

      Abstract:The security of the database system depends not only on the security of the database itself, but also on the security of the network environment and the operating system. The mimicry defense is proposed firstly by Academician WU Jiang-Xing based on the dynamic heterogeneous redundancy architecture, and the database is dependent on the environment. Security has changed from passive defense to active defense. This study uses Mycat as a database middleware, through the separated read and write of the database, the cluster's high availability, distributed transaction processing, fingerprinting SQL identification, to alleviate the data access and processing pressure of a single database. Mycat turns the database module into a heterogeneous dynamic redundancy mode, which enables database mimicking, enabling efficient, fast, and secure access and segmentation of data.

    • Application of Weighted Combination Model Based on XGBoost and LSTM in Sales Forecasting

      2019, 28(10):226-232. DOI: 10.15888/j.cnki.csa.007091

      Abstract (2219) HTML (5160) PDF 1.19 M (2445) Comment (0) Favorites

      Abstract:Aiming at the multi-variable commodity sales forecasting problem, in order to improve the accuracy of prediction, an ARIMA-XGBoost-Lstm weighted combination method is proposed to predict the sales sequence of commodities with multiple influencing factors, In this study, ARIMA is used for univariate prediction. The predicted value is used as a new variable together with other variables in the XGBoost model for mining different attributes, and the predicted values of XGBoost are merged into the multivariate sequence, and then the new multidimensional data is converted. In order to supervise the learning sequence and use the LSTM model for prediction, the three model prediction results are weighted and combined, and the best combination weights are obtained through multiple experiments to calculate the final prediction value. The data results show that the multivariate prediction method based on the weighted combination of XGBoost and LSTM is more accurate than the prediction obtained by a single prediction method.

    • Less-Effort Collision Avoidance for Virtual Pedestrian

      2019, 28(10):233-238. DOI: 10.15888/j.cnki.csa.007076

      Abstract (1331) HTML (706) PDF 1.37 M (1354) Comment (0) Favorites

      Abstract:Real-time multi-agent collision avoidance for large environments with hundreds or thousands of agents need powerful collision avoidance module. Most velocity-obstacles-based method for collision avoidance assume that every agent share the same responsibility to adjust their velocity to avoid potential collision. In order to improve the quality of dynamic collision avoidance for virtual pedestrian simulation, this study uses adjustment factor to distinguish the strategies of different type of pedestrian. And we introduce the less-effort to discuss the relationship between the velocity change and instantaneous energy consumption during dynamic collision avoidance. At last, we use linear programming to choose the best velocity from feasible velocity constructed by the improved ORCA algorithm. The experiment result shows that our method can improve the simulation efficiency of dynamic collision avoidance for large-scale crowd simulation, and the performance also meets the requirements of real-time simulation.

    • Research on Extraction Method of System Log Template

      2019, 28(10):239-244. DOI: 10.15888/j.cnki.csa.007112

      Abstract (1418) HTML (2265) PDF 993.69 K (1848) Comment (0) Favorites

      Abstract:Extracting log template is a very effective way to handle massive system logs. In this study, the Web system log is used as the entry point, extracts the log template by using signature tree model. Based on it, we studied and improved the log preprocessing and template expression generation methods. Aiming at the complex structure problem of syslog, the preprocessing method based on text similarity is adopted to realize the classification of log messages. We used the max template matching method to solve the low template matching problem caused by the inconsistent log format and word-cutting. Finally, we evaluate the experiment of this log template extraction method. The results show that the accuracy of the method is 96.4%, and the template matching degree is greatly increased.

    • Study on Emotional Tendency of New Words Based on Word Vector

      2019, 28(10):245-250. DOI: 10.15888/j.cnki.csa.006879

      Abstract (1164) HTML (713) PDF 939.22 K (1368) Comment (0) Favorites

      Abstract:In order to more specifically express the emotional meaning and tendency of social new words, this study proposes a new word sentiment orientation analysis method based on word vector. In the everchanging development of the information age, due to the continuous development of language application scenarios and the enrichment of extended semantic expressions, many new words expressing emotions appear on the Internet, while the expression of these new words has rich meaning but lacks accurate definition. Therefore, it is difficult to analyze its sentiment orientation. Based on the analysis of the new word discovery method and the word vector training tool Word2Vec, this study focuses on the feasibility and architecture design of the new word orientation analysis method based on Word2Vec, and conducts experiments for microblog corpus. The results show that new words can analyze their emotional tendencies from similar words.

    • Research and Optimization of Routing Algorithm in ZigBee Network

      2019, 28(10):251-256. DOI: 10.15888/j.cnki.csa.007047

      Abstract (1122) HTML (1294) PDF 1.28 M (1589) Comment (0) Favorites

      Abstract:At present, ZBR algorithm is widely used in ZigBee network, but after research and analysis, the energy consumption of ZBR algorithm can be greatly reduced. Therefore, an improved hierarchical energy control algorithm is proposed in this study. The improved algorithm limits the spread range of RREQ packets by controlling the energy threshold of nodes, limits the depth of the network, discards unwanted RREQ packets and reduces them. The energy consumption of the network is compared with the original algorithm in terms of end-to-end delay, residual energy and packet delivery rate through NS-2 simulation experiments. The experimental results show that the improved algorithm can ensure the stability of the network transmission, reduce the delay and energy consumption, and maximize the network lifetime.

    • Decision Tree Algorithms for Lung Cancer Diagnosis Based on Electronic Medical Record

      2019, 28(10):257-263. DOI: 10.15888/j.cnki.csa.007111

      Abstract (3357) HTML (811) PDF 1.16 M (1472) Comment (0) Favorites

      Abstract:With the continuous improvement of people's living standards, the number of cancer diseases is increasing. Among them, lung cancer is a major disease that seriously endangers human health in the 21st century. This paper presents a decision tree method for lung cancer diagnosis based on electronic medical records. Firstly, the characteristics of lung cancer electronic medical records and the instability and over-fitting of the model tree in the decision tree are analyzed. The optimal decision tree model constructed by principal component analysis combined with C5.0 algorithm is used. Firstly, two methods of feature dimension reduction with principal component eigenvalue greater than 1 and principal component cumulative contribution rate greater than 85% are established. Then, the decision tree model and pruning operation are established by C5.0 algorithm. Finally, the data preprocessing process and model are given. The experimental results show that the improved algorithm has better accuracy and good scalability, which proves that the improved algorithm is of great significance for the clinical trial of lung cancer.

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