Loading...
  • 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

Current Issue

    • Development and transplantation of BSP
      based on Godson aerospace chip
      XU Shuai1,2,LIN Bao-jun2,3,4,LIU Ying-chun2,3,ZHAO Shuai3
      2020, 42(04): 571-579. doi:
      Abstract ( 221 )   PDF (1012KB) ( 279 )     
      With the Beidou-3 global navigation satellites system starting to operate in network, the on-board system requires faster speed of data transmission and data operation for the on-board computer systems. The core components of on-board computers for Beidou-3 satellites developed by CAS are all made or autonomously controlled by China. The on-board computers adopt the Godson 1E high-performance aerospace-class processing chip produced by Godson Corporation as the main hardware environment, and the real-time operating system VxWorks as the software environment, which constitute an on-board computer system with high performance of data computing and data transmission. In order to adapt to the new Godson 1E chips, this paper not only develop Board Support Package (BSP) and serial port drivers, but also configure the VxBus-type driver architecture on device driver management, in order to realize the porting and operation of  VxWorks on the new chips. Meanwhile, the reliability, portability, and independence of the driver are effectively improved.
       
      An intra-cluster local-priority efficient-access
      switch in distributed Cachee
      LIU You-yao,ZHANG Yuan,SHAN Rui
      2020, 42(04): 580-587. doi:
      Abstract ( 116 )   PDF (1125KB) ( 181 )      Review attachment
      Reconfigurable array processor has the characteristics of large amounts of memory data, high data parallelism, less global data reuse and obvious data locality. Aiming at these characteristics, this paper proposes an intra-cluster local-priority efficient-access switch in distributed Cache. The switch can make 4×4 PEs to access 4×4 Caches in parallel. Xilinx’s ZYNQ series chip XC7Z045 FFG900-2 FPGA are used for FPGA synthesis. The switch can support the concurrent read and write operations of 16 intra-cluster PEs in the absence of conflicts, the working frequency can reach 221 MHz, and the memory bandwidth can attain 7.6 GB/s. The image texture extraction algorithm based on Gray-level Co-occurrence Matrix (GLCM) is implemented on this switch. The data memory bandwidth reaches 478.125 MB/s, and the execution time is 0.24ms.
       
      Diagnosability and a new diagnosis algorithm
      of Cross-cube in fault situation
      WANG Xi1,2,ZHANG Shu-kui2
      2020, 42(04): 588-595. doi:
      Abstract ( 106 )   PDF (1752KB) ( 157 )      Review attachment
      Supercomputers based on parallel systems have always been a hot research topic in academia and industry. As the basis of parallel systems, the properties of interconnection network determine the system performance directly. As a variant of hypercube, cross-cube is a significant interconnection network, which has superior properties such as low diameter compared with hypercube. This paper uses PMC diagnostic model and graph theory method to study the exact value of the diagnostic degree of cross-cube in fault situation. Then, a diagnosis algorithm is proposed and its time complexity is analyzed. Furthermore, simulation experiments verify that the diagnostic algorithm has higher efficiency than some literature algorithms under a variety of fault parameters. Our research can measure the reliability of cross-cube more accurately.
       

       

      Pareto dominance based area and power
      consumption optimization of MPRM circuit
      YAN Pan-pan,YU Hai-zhen,SHI Xu-hua,WAN Kai
      2020, 42(04): 596-602. doi:
      Abstract ( 100 )   PDF (531KB) ( 133 )      Review attachment
      Aiming at the comprehensive optimization problem of area and power consumption of MPRM circuits, an optimal polarity search scheme, called Multi-Objective Ternary Diversity Particle Swarm Optimization (MOTDPSO), is proposed. On the basis of Ternary Diversity Particle Swarm Optimization (TDPSO) solving the comprehensive optimization problem of MPRM circuits, the mutation operator is introduced to perturb the particle, and the boundary constraint processing is performed on the particle beyond the defined boundary range. The concept of Pareto dominance is used to improve the algorithm. Then, the parameter mapping relationship between the particle based on Pareto dominance and the polarity of MPRM circuits is established. By combining the area and power evaluation model and the OR/XNOR circuit mixed polarity conversion method, the algorithm is applied to optimize the area and power consumption of MPRM circuits. Finally, tests on 18 MCNC benchmark circuits with PLA format show that, compared with the NSGA-II algorithm, the average area optimization rate of the optimal solution obtained by the MOTDPSO algorithm is 4.29%, and the average power consumption optimization rate is 6.02%.
       
      A multi-objective optimization algorithm without
      parameter grouping and with large-scale variables
      ZHU Deng-jing,DUAN Qian-qian
      2020, 42(04): 603-609. doi:
      Abstract ( 112 )   PDF (521KB) ( 149 )      Review attachment
      At present, most multi-objective optimization algorithms do not consider the interaction between decision variables, but just optimize all variables as a whole. With the increase of decision variables, the performance of multi-objective optimization algorithms will decrease sharply. Aiming at the above problems, a multi-objective optimization algorithm without parameter grouping and with large-scale variables (MOEA/DWPG) is proposed. By combining collaborative optimization with the decomposition-based multi-objective optimization algorithm (MOEA/D), this algorithm designs a grouping method without parameters to improve the grouping accuracy of interaction variables and to enhance the algorithm performance when it handles the multi-objective optimization problems with large-scale variables. Experimental results show that the proposed algorithm is significantly superior to MOEA/D and other advanced algorithms on the multi-objective problems with large-scale variables.
       
      Research on Android ransomware protection technology
      HU Jian-wei,ZHANG Yu,CUI Yan-peng
      2020, 42(04): 610-619. doi:
      Abstract ( 120 )   PDF (613KB) ( 189 )     

      With the popularity of smart devices such as smartphones and pads, the attack of Android ransomware is becoming increasingly serious. Compared with other malicious software, ransomware is widely favored by hackers because of hard restoration and the directness of obtaining benefits, which also brings serious spiritual and property damage to users. To protect our smart devices from ransomware and reduce threats and losses, researchers conduct a lot of research on Android ransomware and propose many practical detection schemes. This paper first summarizes the characteristics of Android ransomware, and then summarizes the existing research work on detecting and safeguarding against ransomware on the Android platform and makes a comprehensive analysis and comparison on them. Finally, we point out the remaining problems of these solutions, put forward corresponding suggestions and discuss future research directions.

      An energy-balanced WSNs routing optimization
       algorithm based on AGNES clustering
      GOU Ping-zhang,ZHANG Fen,MAO Gang,JIA Xiang-dong
      2020, 42(04): 620-627. doi:
      Abstract ( 121 )   PDF (717KB) ( 216 )     
      The life cycle of a wireless sensor network is directly related to the energy consumption of its nodes. In order to solve the problem that the uneven distribution of energy consumption affects the network life, an energy-balanced WSN routing optimization algorithm (EBRAA) based on AGNES clustering is proposed. The AGNES clustering algorithm is used to obtain uniform clustering of the network. According to the residual energy of the nodes in the cluster, the distance between the nodes and the base station, and their weights, the distributed cluster heads are selected. The improved Dijkstra algorithm is used to generate the multi-hop routing with the shortest path between the cluster heads. The simulation results show that, compared with LEACH and KBECRA algorithms, EBRAA algorithm has more reasonable cluster distribution and more balanced energy consumption, and prolongs the network life cycle.
       
      Dynamic mining of sensitive data streams in
      heterogeneous complex information networks
      XIONG Ju-xia1,2,3,WU Jin-zhao1,2,3
      2020, 42(04): 628-633. doi:
      Abstract ( 88 )   PDF (758KB) ( 128 )     
      For the sensitive data streams with high-dimensional redundancy in heterogeneous complex information networks, the probability of data feature formation is low, which leads to multiple mining, high memory usage, low mining accuracy and long running time. Aiming at the above problems, a dynamic network sensitive data stream mining method based on the maximum inter-class divergence is proposed. The maximum difference interval between sensitive data is used as the basis for classification to obtain the maximum inter-class divergence of the network sensitive data. The optimal divergence iterative function is determined in the genetic iterative state. The mining characteristics of the iterative function are preferably selected to obtain the dynamic mining characteristics. Clustering analysis is performed on the mining characteristics to obtain data hiding information modes. These modes are evaluated, and knowledge representation is carried out on the reasonable information modes, so as to realize the dynamic mining of the sensitive data streams in the heterogeneous complex information networks. The experimental results show that the mineable feature formation probability of the method can be up to 98%, and the labels are close to the actual values. The method has the advantages of high mining accuracy, short running time and low memory usage.
       
      A configuration constraint extraction
      method for enumeration type
      ZENG Guang-fu,HE Hao-chen,ZHOU Shu-lin
      2020, 42(04): 634-640. doi:
      Abstract ( 113 )   PDF (777KB) ( 154 )     
      It is reported that software configuration failure has become an important factor causing computer system anomalies and crashes. Configuration failures are caused by users' misconfigurations due to their inability to adequately obtain configuration constraint information. Because users lack software domain knowledge, configuration failures are difficult to avoid. Therefore, how to accurately analyze and extract the constraints of software configuration items to provide a basis for software configuration fault diagnosis and repair has important research significance. Specifically, enumeration is a common type of software systems, and the limitation of its value space often causes the system software configuration failure. This paper systematically investigates the enumeration type configuration constraints of six commonly used C/C++ open source software including Apache Httpd, Nginx, Postfix, MySQL, Redis, and PostgreSQL. Aiming at the problem that the value spaces of enumeration type configuration items are underreported in the previous methods, based on the program analysis method, an automatic configuration constraint extraction method for enumeration type configuration is designed and implemented. This method greatly improves the accuracy of the configuration constraint extraction of the above open source software, and improves the availability of the software configuration, and the diagnosibility of configuration failure.
       
      A generalized dynamic integrated neural network
      model based on fault-correction waiting delay
      HUI Zi-qing,LIU Xiao-yan,YAN Xin
      2020, 42(04): 641-648. doi:
      Abstract ( 88 )   PDF (812KB) ( 117 )      Review attachment
      The software reliability growth model plays an important role in reliability evaluation and guarantee. Aiming at the problems of fault detection and fault-correction waiting delay in software testing, this paper proposes a generalized dynamic integrated neural network model considering the fault-correction waiting delay. The model considers the diversity of software engineering. It uses the neural network method to construct a generalized dynamic integration model, and considers the fault-correction waiting delay phenomenon to complete the fault detection and prediction. Through the experiments on two real failure datasets (DS1 and DS2), the proposed method is compared with the existing software reliability growth model. The results show that the neural network model considering the fault-correction waiting delay has the best fitting effect, and exhibits better software reliability assessment performance and model versatility.
       
       
       
      A small object detection algorithm based
      on deep convolutional neural network
      LI Hang1,2,ZHU Ming1
      2020, 42(04): 649-657. doi:
      Abstract ( 212 )   PDF (1522KB) ( 307 )      Review attachment
      In view of the shortcomings of YOLO object detection algorithm in small object detection, and the difficulty of achieving real-time performance on embedded platforms, this paper designs an improved YOLO object detection algorithm, called dense_YOLO. The algorithm contains two phases: feature extraction phase and object detection regression phase. In the feature extraction phase, based on the idea of DenseNet structure, a new slim-densenet feature extraction module based on deep separable convolution is designed, which enhances the transmission of small object features and reduces the parameter quantity to accelerate the network propagation speed. In the object detection stage, the idea of adaptive multi-scale fusion detection is proposed to fuse the extracted features, and the objects are classified and regressed on different feature scales, which improves the detection accuracy of small objects. Experimental results show that, compared with the original YOLO object detection algorithm, the dense_YOLO object detection algorithm improves mAP by 7%, decreases the single picture detection time by 15 ms, and reduces the model size by 90 MB.
       
      Facial feature point localization based on C-Canny
      algorithm and improved single neural network
       
      FU Wen-bo1,2,HE Xin1,2,YU Jun-yang1,2
      2020, 42(04): 658-664. doi:
      Abstract ( 105 )   PDF (763KB) ( 129 )      Review attachment
      Deep learning has achieved remarkable results in the field of facial recognition. However, when dealing with facial images under complex conditions such as occlusion, illumination and improper angles, predicting a large number of facial feature points is still a challenging problem. In order to solve the localization problem of multiple facial feature points under complex conditions, this paper designs a network structure based on C-Canny algorithm and improved single neural network. The traditional Canny algorithm is applied to the face region localization stage, so that the neural network can quickly reposition the face region to improve the accuracy of model recognition. Experimental results show that, compared with some existing traditional algorithms and neural networks, the neural network structure reduces the value of loss function by 12.2% on average on the 300-w and 300-vw datasets.
       
      A multi-target tracking algorithm based on YOLO detection
      LI Xing-chen,LIU Xiao-ming,CHENG Xiao-nan
      2020, 42(04): 665-672. doi:
      Abstract ( 336 )   PDF (922KB) ( 290 )     
      Aiming at the occlusion problem in the current video multi-target tracking process, a multi-target detection and tracking algorithm combining YOLO v3 is proposed. The framework based on detection tracking is selected as the overall framework for tracking, and YOLO v3 is used to detect the target information. Based on the selected detection category, the proposed tracking algorithm is used to complete the multi-target tracking of this category through data association. Aiming at the problems of target occlusion in the tracking process and abnormal trajectory tracking caused by target occlusion, a correction algorithm is proposed. Most of the occluded targets in the test video can be accurately tracked in the experiment, but some target identity swaps will occur when the background moves. The proposed algorithm has certain accuracy and real-time performance in solving the occlusion problem in multi-target tracking.
       
      Defect detection of mobile phone
      motherboard based on RetinaNet
      MA Mei-rong,LI Dong-xi
      2020, 42(04): 673-682. doi:
      Abstract ( 152 )   PDF (1012KB) ( 269 )      Review attachment
      Since the motherboard images of different mobile phones have a multi-resolution imaging mode, the shape of defective components become multi-scale. Conventional defect detection methods mainly include image fusion methods or statistical model extraction methods, but the robustness of these methods still needs to be improved. To solve this problem, an automatic learning representation model, called RetinaNet object detector, is proposed. Firstly, feature pyramid network (FPN) is used to extract the multi-scale feature classification and location of defective components, and MobileNetV2 is introduced to compress and accelerate the RetinaNet model. Secondly, focus loss is used to resolve the class imbalance and increase the difficulty of detecting the contribution of the samples to the loss during the training. The experimental results show that RetinaNet can effectively detect defective components of different scales, and has high detection accuracy. Compared with other object detectors, RetinaNet achieves an average accuracy (mAP) of over 95%. These results demonstrate the effectiveness of the proposed model.
       
      Road extraction algorithm based on
      prediction and residual refinement networks
      #br#  
      XIONG Wei1,2,GUAN Lai-fu1,WANG Chuan-sheng1,TONG Lei1,LI Li-rong1,LIU Min1
      2020, 42(04): 683-690. doi:
      Abstract ( 67 )   PDF (834KB) ( 130 )     
      Aiming at the problem of road detection in aerial images, a road extraction algorithm based on prediction and residual refinement networks is proposed. Firstly, the prediction network makes initial predictions. In order to improve the refinement ability of the segmentation network and learn higher- level road features, the dilated convolution and multi-kernel pooling modules are introduced in the prediction networks. Secondly, the residual refinement network will further refine the output of the prediction network and improve the ambiguity of the prediction network results. In addition, considering the small proportion of road pixels in aerial images, the network also combines binary cross entropy, structural similarity, and intersection over union loss functions to reduce road information loss. The experimental results on the Massachusetts road dataset show that the precision, recall, F value and accuracy reaches 99.3%, 95.7%, 97.3% and 95.1%, respectively. The intersection over union and structural similarity also reaches 94.8% and 84.3%, respectively. Compared with other algorithms, this proposed algorithm has certain application value.

       

       

       
      Video analysis of subway tunnel inspection
       based on deep separable convolution
      SUN Ming-hua,YANG Yuan,LI Yuan-bo
      2020, 42(04): 691-698. doi:
      Abstract ( 112 )   PDF (783KB) ( 158 )     
      Currently, the safety of subway tunnels mainly relies on the manual track inspection of subway track inspectors when there are no trains on the track. This method is slow and inefficient, and the inspection results are completely dependent on the experience and status of the track inspector. Aiming at this problem, a video anomaly alarm system of subway tunnel inspection based on deep separable convolution is proposed, this system uses the proposed SubwayNet convolutional neural network to complete the classification of inspection video images. The built-in convolutional neural network is trained and saved by using the produced subway tunnel inspection dataset. The graphical user interface is created and the alarm function is added. Finally, the program files are packaged into an executable file. The experimental results show that the classification accuracy of the system can reach 96%, and the speed can reach 52 frames/second, which meets the requirements of real-time and accurate analysis of video.
       
      Image style migration based on
      cycle generative adversarial networks
       
      PENG Yan-fei,Wang Kai-xin,Mei Jin-ye,SANG Yu,ZI Ling-ling
      2020, 42(04): 699-706. doi:
      Abstract ( 186 )   PDF (849KB) ( 188 )     
      Image style migration refers to learning the style of oil painting pictures and applying the learned style to other pictures to make the pictures have the style of oil painting. The current methods based on generative adversarial networks have been widely used in image style migration. Aiming at the problem that Cycle Generative Adversarial Networks (CycleGAN) do not have high texture definition when processing images, a method of adding a Local Binary Pattern (LBP) algorithm is proposed. The LBP algorithm is added into the generation model of CycleGAN to enhance the extraction of image texture features by CycleGAN. Aiming at the problem of noise in the generated images, Total Variation Loss is added into the loss function to constrain the noise. The experimental results show that the quality of the generated images can be improved by adding LBP algorithm and Total Variation Loss, and the generated images have better visual effects.
       
      Object detection in traffic scenes based on anchor-free
      GE Ming-jin,SUN Zuo-lei,KONG Wei
      2020, 42(04): 707-713. doi:
      Abstract ( 147 )   PDF (761KB) ( 232 )      Review attachment
      Object detection using deep learning methods in the field of intelligent transportation has become a research hotspot. Currently, most of the classic object detection algorithms, whether which are the single-stage object detection models based on regression or the two-stage object detection models based on candidate regions, use a large number of predefined priori boxes called “anchor” to enumerate the possible positions, sizes and aspect ratios so as to search the objects. It will cause serious imbalance between positive and negative samples, and the performance and generalization ability of the models are also limited by the anchor's design. Aiming at the above problems of the anchor-based object detection algorithms, a single-stage object detection network, called RetinaNet, is used to establish the anchor-free based object detection models for vehicles, pedestrians, and cyclists in traffic scenes. Pixel-by-pixel prediction is adopted to handle object detection and add central prediction branches to improve the detection performance. Experiments show that, compared with the original RetinaNet algorithm based on anchor, the improved algorithm can better recognize vehicles, pedestrians, and cyclists in traffic scenes.
       
      null
      YUAN Quan1,2,3,CHENG Zhen-hua1,2,JIANG Yang1,2
      2020, 42(04): 714-721. doi:
      Abstract ( 316 )   PDF (669KB) ( 273 )     
      null
      An improved density peak algorithm for
      micro-learning unit text clustering based on LSA model
      WU Guo-sheng, ZHANG Yue-qin
      2020, 42(04): 722-732. doi:
      Abstract ( 107 )   PDF (1226KB) ( 141 )     
      With the explosive growth of micro-learning resources, a large number of unprocessed fragmented text resources bring great inconvenience to learners. In order to help learners to find suitable contents from fragmented resources for personalized learning, it is necessary to cluster micro-learning resources in the form of text. Therefore, this paper attempts to apply an improved density peak algorithm to micro-learning unit text clustering. Aiming at the problems of high dimensional sparse vector space, insufficient global consistency, cutoff distance sensitivity, and supervised selection of density peak centers when the density peak algorithm perform clustering in its field, this paper proposes two approaches based on Latent Semantic Analysis (LSA) model. Firstly, a new definition of local density is proposed according to clustering requirements, density sensitive distance is used as the clustering criteria, and the global consistency problem of clustering is solved by solving the problem of cutoff distance sensitivity. Secondly, outliers are found by linear fitting to automatically find the density peak centers in order to realize unsupervised selection problem of peak centers. Experimental results on real data sets of micro-learning units show that the proposal is more suitable for text clustering of micro-learning units than the original algorithm and other classical clustering algorithms.