Computer and Modernization

Detection and Tracking of Hard Hat Wearing Based on Deep Learning

QIN Jia, CAO Xue-hong, JIAO Liang-bao

2020, 0(06): 1.

Asbtract ( 445 )

PDF (2454KB) ( 223 )

References | Related Articles | Metrics

In order to solve the shortcomings of traditional construction site safety management and reduce casualties caused by construction workers not wearing hard hats, a method for detecting and tracking hard hat wearing based on deep learning is proposed. Firstly, the YOLOv3 target detection network is used to realize the helmet wearing detection, and the Kalman filter and the KM algorithm are used to implement multi-target tracking and counting. The test results at a complex construction site show that the detection speed of the network model can reach 45 fps, with an average accuracy of 93%, and the accuracy and recall rates without a helmet are 97% and 95% respectively. This model basically realizes the real-time detection of the wearing condition of the helmet.

Target Detection Algorithm of Gaussian Mixture Model Combined with HED Network and Double Threshold Segmentation

LI Rui, WANG De-zhong

2020, 0(06): 7.

Asbtract ( 239 )

PDF (2639KB) ( 104 )

References | Related Articles | Metrics

In the process of target detection, the GMM is easily interfered by lighting, the similarity of target color and background color, target shadow and shooting height. Aiming at the above problems, a GMM algorithm is combined with improved HED network and OTSU double threshold segmentation is proposed. First, the improved model divides the background, noise and foreground targets of video frame by double threshold, and reasonably selects the number of GMM. Secondly, HED network is used for edge detection of the input images.The “and” operation of edge result detected by the HED network and the GMM detection result of the double threshold segmentation is completed to obtain the final target detection result. Experimental results show that the improved algorithm has a higher detection rate, a more complete detection profile “and” a better detection effect.

A Multi-classification Approach to Discriminate Causes of Packet Losses and Errors for High Throughput Wireless LAN Based on Convolution Neural Network

ZHANG Ning, HUANG Ting-pei, CAI Li-ping, LI Shi-bao

2020, 0(06): 14.

Asbtract ( 192 )

PDF (2429KB) ( 91 )

References | Related Articles | Metrics

Channel errors and collisions are two major factors that cause packet loss and errors in wireless networks. The reason for effectively identifying packet loss and errors is the basis for implementing the high performance IEEE 802.11n protocol. This paper focuses on how to improve the accuracy of discriminating the causes of packet losses and errors with low overhead. Based on the supervised learning theory, this paper proposes a light-weighted discriminator, named MPLEC, to differentiate the root causes of packet losses and errors with high accuracy and timeliness. MPLEC analyzes the packet reception situation through a large number of field scene statistical experiments, extracts the feature vector composed of RSSI and CSI as the input of the supervised learning model, and performs offline training and inspection on the multi-class MPLEC classification model through supervised learning method. The result indicates that the accuracy rate of MPLEC is as high as 87%. Finally, this paper applies the MPLEC to the CSMA/CA protocol to evaluate its performance. The experimental results show that compared with the original backoff algorithm, this method can increase the probability of successful retransmission by 25% and increase the time utilization by 7.8%.

A Hierarchical Cooperative Caching System in Mobile Streaming Media

YANG Da-wu, LI Ze-ping

2020, 0(06): 22.

Asbtract ( 209 )

PDF (961KB) ( 90 )

References | Related Articles | Metrics

With the development of mobile Internet and the increase of the number of users, dynamic caching mechanism is widely used in audio & video services to reduce the data traffic going through backhaul backbone links and improve the viewing experience. How to adjust the cache content of multiple nodes according to the network and user demand to cut down the traffic in backbone links is an important problem to be solved in current cache deployment. Based on the submodular function theory, the proactive and reactive resource allocation adjustment schemes and their algorithms are proposed. According to the popularity of the content, the proactive scheme puts the video files to the cache space to minimize overhead of obtaining them. The reactive one adjusts the content in the space of the node in time to cater to the popularity’s change, so as to improve the utilization of the cache resources and user experience, and reduce the rate of occupied bandwidth in the backbone links. The complexity of the minimum access cost algorithm is related to the size of the cache space, which can quickly iterate out the resource allocation scheme when the cache space is tight. Numerical simulation shows that the proactive and reactive resource allocation schemes can effectively alleviate the traffic load of the remote server and improve the user’s experience.

Reliability Analysis Method for Repairable Distributed System in Cloud Compute Environment

YANG Mu-chuan, LYU Xiao-dan, JIANG Chao-hui,

2020, 0(06): 28.

Asbtract ( 261 )

PDF (1483KB) ( 172 )

References | Related Articles | Metrics

With the further development of cloud computing technology, more and more application systems are hosted on cloud computing platforms, which puts forward higher requirements for the reliability of the many distributed systems that make up a cloud computing platform. It is difficult for traditional analysis methods to analyze the reliability of repairable distributed system when the system scale is large and dynamic. In order to improve service quality and reduce economic losses caused by violation of service level agreements, this paper proposes a reliability analysis method for repairable distributed systems based on Markov models. By simplifying the state space of the system, the software and hardware states are sampled during the system operation, and the failure process and repair process of the distributed system are analyzed. According to the failure probability sequence and repair probability sequence in a given time, the node state transition matrix of the distributed system is calculated, and the steady-state vector corresponding to the Markov matrix is obtained. Then according to the characteristics of the distributed system, the steady-state vector is further analyzed to obtain the final reliability measurement index of the system. Finally, the validity and effectiveness of the method are verified by experiments.

Safety Analysis of Three Classic Blockchain Consensus Mechanisms

DENG Ming-wei, OU Wei, YANG Jie

2020, 0(06): 34.

Asbtract ( 399 )

PDF (901KB) ( 179 )

References | Related Articles | Metrics

Blockchain technology originated from the emergence of bitcoin. With the popularity of bitcoin and other digital currencies in the financial market, blockchain technology has attracted wide attention from all circles. The consensus mechanism is the core of blockchain technology, and the ledger and data on the blockchain are the products of the consensus mechanism. Once the safety of the consensus mechanism is in question, the availability and credibility of blockchain will be seriously hit. Starting from the security of the consensus mechanism, this paper first introduces the concepts of blockchain and consensus mechanism, and then makes a comparative analysis of three classic blockchain consensus mechanisms. By studying the attack methods of the three consensus mechanisms one by one, their consistency and adversary models are compared. Finally, the limitations of current consensus mechanism safety analysis are pointed out, and the direction of future research on consensus mechanism safety is determined.

Design of Parking Lot Path Guidance System Based on Multi-objective Point A* Algorithm

XIAO Wei, ZHANG Lei , QIU Ze-hua, ZHONG Yong, ZHANG Tie-nan

2020, 0(06): 40.

Asbtract ( 268 )

PDF (1005KB) ( 126 )

References | Related Articles | Metrics

Aiming at the problem that most parking lots lack a parking guidance system which is easy to implement in engineering, which makes it difficult for users to parking and has some potential safety hazards, a parking guidance system based on the multi-objective point A* algorithm is designed. The system is composed of four parts: entrance and exit control module, parking status monitoring module, path guidance module and information processing center. It can be based on the existing parking infrastructure to the greatest extent, and has the characteristics of small amount of engineering and low cost. The designed multi-objective point A* algorithm can select an optimal target point from multiple potential target points and complete the optimal path planning. Finally, taking a residential parking lot as an example, the system is simulated by using Microsoft Visual Studio 2015 platform, and three common modes of parking guidance, including the latest parking guidance mode, the least walking guidance mode and the continuous free parking guidance mode, are emphatically run. The simulation results verify the effectiveness and feasibility of the system.

Preprocessing Technology of Gas Drainage Data and Application of Criteria for Achieving Standards

WU Ke-jie, XU Jin, CHEN Qing, ZHANG Yi

2020, 0(06): 46.

Asbtract ( 164 )

PDF (886KB) ( 94 )

References | Related Articles | Metrics

Aiming at the problems of long period of manual measurement data and data missing of on-line monitoring device for gas extraction, the relationship between gas extraction time and cumulative extraction volume of underground pipeline and the repair algorithm of gas monitoring data in two cases of single point missing and continuous missing are studied by data fitting method, and a set of gas monitoring data repair algorithm is designed and implemented. Evaluation system of gas drainage meets the standard. The test results of the extraction evaluation unit of N1633 North return air roadway in Shigou Coal Mine show that the system meets the evaluation requirements of the gas extraction standard, and provides an application demonstration for the gas extraction standard at the present stage of the mine.

Automatic Generation Method of Ada Code for Aerospace Embedded Software Based on AADL

FENG Si-zhe, YANG Zhi-bin, XUE Lei

2020, 0(06): 52.

Asbtract ( 206 )

PDF (1998KB) ( 103 )

References | Related Articles | Metrics

Model-Driven Development (MDD) is gradually applied to the design and implementation of safety-critical software in aerospace and other fields.Architecture Analysis and Design Language (AADL) is a standardized embedded software architecture description language that provides complete support for the design and implementation of safety-critical software through modeling, verification and code generation.However, the code in the industry runs on the target platform with different characteristics, such as different hardware and software architectures and programming interfaces. The existing researches on AADL code generation mainly integrate the automatically generated code into the platform manually, which is tedious and error-prone. This paper presents an automatic generation method of Ada code for aerospace embedded software based on AADL.Firstly, the AADL modeling of satellite attitude and orbit control system is given.Secondly, the automatic transformation rules of Ada code related to AADL to platform are given. Finally, a prototype tool for code generation is given, and the code generated by the AADL model of the satellite attitude and orbit control system is checked by the space coding standard, and the effectiveness of the method proposed in this paper is verified by running in the relevant simulation environment.

A Symbolic Execution Technology Supporting Multi-thread Program

LI Tong, DING Guo-fu

2020, 0(06): 60.

Asbtract ( 266 )

PDF (1290KB) ( 112 )

References | Related Articles | Metrics

Symbolic execution is a useful method in program verifying. But most symbolic execution tools don’t support multi-thread program. This paper analyzes the existed tools which support multi-thread program, and finds these problems: 1) some tools have good performance, but do not support lib, and hard to use practically; 2) some tools support lib, but are too old to upgrade, and can not find some basic bugs such as subtraction overflow, multiplication overflow and shift overflow. To solve these problems, this paper designs and implements MTSE(Multi-Thread Symbolic Execution) based on KLEE. MTSE supports multi-thread program with libc and libc++. MTSE can find 50% more bugs than Cloud9, and bring about 30% improvement in both instruction coverage and branch coverage compared with Cloud9.

Application of Algorithm of Fast Mining Maximal Frequent Itemsets in Library Management

YU Hai-yang

2020, 0(06): 68.

Asbtract ( 213 )

PDF (821KB) ( 132 )

References | Related Articles | Metrics

Aiming at the lag of librarys service mode and the contradiction between librarys supply and users demand, this paper uses data mining technology and DS-Eclat algorithm to mine its maximum frequent item set by borrowing records, and to promote the transformation of librarys service mode by finding out the internal association rules in users searching information.By comparing the traditional Eclat algorithm with the DS-Eclat algorithm in this paper, it is shown that the DS-Eclat algorithm can quickly discover the maximum frequent item set, and the maximum frequent item set can promote the development of library personalized service.

High-dimensional Numerical Anomaly Data Detection Based on Multi-level Sequence Integration

LI Ke-xin, LI Jing, SHAO Jia-wei, XIAO Yi

2020, 0(06): 73.

Asbtract ( 187 )

PDF (1530KB) ( 78 )

References | Related Articles | Metrics

With the rapid development of big data, data analysis and knowledge discovery have become research hotspots, and anomaly data detection is the key to data quality improvement. The abnormal data detection method based on sequence ensemble learning may cause large deviations in the detection of abnormal data in high-dimensional numerical data due to noise data and excessive number of dimensions. This paper proposes a high-dimensional numerical anomaly data detection model of multi-layer sequence ensemble learning based on elastic network. Each layer contains three modules: abnormal data candidate set module, elastic network dimension reduction module and data abnormality scoring module. First, the abnormal data candidate set selection module selects some possible abnormal data according to abnormal score. Then, the elastic network reduces the dimension of data according to the outlier candidate set and its abnormal score. Finally, the selected features related to the abnormal score are used to score the data again. The threshold in each layer of the abnormal data candidate set selection module is set to a different value, and each layer is executed cyclically until the mean square error of the current elastic network is greater than the previous or the current detection precision is smaller than the initial detection precision. In the experimental stage, the high-dimensional anomaly data set provided by ODDS is used to test the performance of the model proposed in this paper based on the detection accuracy, the number of extracted features, the convergence speed, etc. The results show that the proposed method can not only improve the detection accuracy of high-dimensional numerical anomaly data, but also effectively reduce the effect of noise on the detection results.

Comparative Study of Oversampling and Ensemble Learning Methods in Software Defect Prediction

WANG Hai, JIANG Feng, DU Jun-wei, ZHAO Jun

2020, 0(06): 83.

Asbtract ( 261 )

PDF (604KB) ( 137 )

References | Related Articles | Metrics

In recent years, research on software defect prediction has attracted much attention. Class-imbalance problem is common in software defect prediction, that is, the number of defective samples is much smaller than that of non-defective samples, but defective samples are the focus of prediction. Due to the above problem, the performance of defect prediction models is difficult to meet the requirement of users, hence it is necessary to effectively process the imbalanced data. At present, sampling-based methods and ensemble learning methods have become two important methods for dealing with imbalanced data. Many researchers have proposed different oversampling methods and ensemble learning methods. This paper studies how to better combine these two kinds of methods to effectively deal with the class-imbalance problem in defect prediction. For that purpose, this paper selects four common oversampling methods (i.e., RandomOverSampler, SMOTE, Borderline-SMOTE and ADASYN) and four commonly used ensemble learning methods (i.e., Bagging, Random Forest, AdaBoost and GBDT). This paper respectively combines one oversampling method with one ensemble method, and hence forms different combinations. By comparing the defect prediction performance of each combination, the optimal combination is obtained, which may provide some useful insights for the processing of imbalance problem in defect prediction. Experiment results demonstrate that the oversampling method ADASYN has more advantages in dealing with the imbalance problem. The oversampling methed ADASYN and the ensemble method GBDT is the best combination, which has better defect prediction performance than other combinations.

Spam Text Classification Method Based on Deep Q-network

JING Dong-sheng, XUE Jing-song, FENG Ren-jun

2020, 0(06): 89.

Asbtract ( 288 )

PDF (792KB) ( 179 )

References | Related Articles | Metrics

Electronic mail is widely used in people’s daily life. It also serves, however, as a carrier for the proliferation of spam mails filled with false information, malicious software and undesired advertisements. Spam mails not only bring inconvenience but also unnecessarily consume a lot of network resource and even pose a huge threat to their information safety. Therefore, it remains an important task to effectively identify and filter spam mails. Current filtering methods are mainly based on identifying the source and content of mails, which are not effective and require a large amount of artificial labeling and are not sensitive to the changes of spam mails’ content or format. In recent years, researchers have applied deep reinforcement learning to the natural language processing and obtained good results. Therefore, this paper presents a classification method for identifying spam mails based on deep Q-network. The mail text first is preprocessed, then is segmented and is transformed into word vectors using Word2vec model. The deep Q-network is used to filter spam mails based on these word vectors in order to improve efficiency and accuracy. The method makes full use of the CBOW model in Word2vec to obtain the word vector corresponding to each participle in the mail text, and directly processes the obtained word vector with the deep Q-network, without extracting the features of the mail, so as to avoid the negative impact caused by the deviation of feature extraction. The experiment results verify the effectiveness of the method.

Image Captioning Based on Adaptive Attention Model

HOU Xing-chen, WANG Jin

2020, 0(06): 95.

Asbtract ( 237 )

PDF (2141KB) ( 99 )

References | Related Articles | Metrics

Attention-based neural encoder-decoder frameworks have been widely adopted for image captioning. Most methods force visual attention to be active for every generated word. However, the decoder likely requires little to no visual information from the image to predict nonvisual words such as “the” and “of”. In this paper，an adaptive attention model is proposed, in which the encoder adopts the Faster R-CNN network to extract the salient features of images, the decoder LSTM network adapts a visual sentinel. At each time step, it can automatically decide when to rely on visual signals and when to just rely on the language model. Finally, the model is verified on Flickr30K and MS-COCO data sets, the experimental results show that the model effectively improves the quality of image captioning.

Prediction of Health Index Based on Improved Non-equidistant Grey Model

LI Rui, LI Xiao-hui, LI Xiao-yu , DING Yue-min

2020, 0(06): 101.

Asbtract ( 208 )

PDF (789KB) ( 84 )

References | Related Articles | Metrics

Aiming at the problems that it is difficult to predict the future health situation through the finite health data with unequal time interval sampling and the accuracy of the traditional non-equidistant grey prediction model is low in the short-term health prediction, an improved non-equidistant grey Markov prediction model is proposed. Firstly, the improved model reduces the impact of data mutation on the prediction results through data preprocessing and optimization of the prediction process. Secondly, the optimal weight coefficient is designed to optimize the model construction. Finally, the residual error is corrected by the strategy of grey and Markov correction. After the comparative analysis on actual monitoring data, the results show that the proposed model has higher prediction accuracy, so that the short-term health situation can be predicted relatively accurately.

Innovation of Computer Vision Teaching Contents Under Development of Deep Learning

CHEN Chuan, CHEN Zhe, DING Shuang-hui

2020, 0(06): 107.

Asbtract ( 245 )

PDF (1032KB) ( 93 )

References | Related Articles | Metrics

In recent years, deep learning theory and application technology have achieved rapid development, and its application in computer vision has been increasingly extensive and in-depth. It has made remarkable achievements in many computer vision tasks, and brought significant impacts on the existing computer vision teaching content. On the basis of summarizing the application status of deep learning theory in various aspects of computer vision, this paper proposes adaptive innovation of computer vision teaching content, and integrates deep learning theory into computer vision teaching, so as to better reflect the promotion effects of theoretical development of relevant disciplines on the reform of computer vision teaching content.

An Improved Electric Vehicle Charging/Switching Facility Layout Considering Traffic Flow

ZHANG Zhong, ZOU Yan-fei, LIU Shu-ying

2020, 0(06): 114.

Asbtract ( 216 )

PDF (1676KB) ( 80 )

References | Related Articles | Metrics

In order to effectively promote the vigorous development of electric vehicles in China, improve the use efficiency of electric vehicles and reduce the environmental pollution in the transportation and logistics industry, this paper intends to carry out a study on the optimal layout of electric vehicle charging/switching facilities by constructing the planning method of electric vehicle charging/switching facilities that takes into account the traffic flow. Firstly, the layout principles of charging/switching facilities of electric vehicles are analyzed from four aspects: convenience principle, economy principle, safety principle and feasibility principle. On this basis, the limitations of the typical site selection model are comprehensively analyzed. With the goal of maximizing the service capacity of the charging/switching facilities, the site selection model of the intercepting flow of the charging/switching facilities of the advanced electric vehicle is constructed, taking into account the traffic flow. Then, considering the NP-hard characteristics of the model, the binary particle swarm optimization (PSO) algorithm is introduced into the distribution model solution of charging/switching facilities to improve the calculation efficiency. Finally, taking a region of Shenzhen city as an example, the planning and layout study of electric vehicle charging/switching facilities is conducted based on MATLAB software to provide an effective plan for the charging/switching facilities layout in this region.

Improvement of Fuzzy C-Means Clustering Algorithm Based on Self-paced Data Reconstruction Regularization

CHEN Yi-jun, CAO Luo-wei, DU Yu-qian

2020, 0(06): 120.

Asbtract ( 157 )

PDF (1544KB) ( 75 )

References | Related Articles | Metrics

In order to reduce the sensitivity of fuzzy C-means clustering algorithm for outliers and noise data points, a self-paced data reconstruction is proposed. Traditional fuzzy C-means algorithm realizes fuzzification of memberships by introducing a weighting parameter into the objective function of the C-means clustering. This paper achieves fuzzification of memberships through regularization of hard C-means clustering by data reconstruction. In addition, the proposed algorithm gradually carries out the clustering of data points in a self-paced manner. Experimental results show that the algorithm can significantly reduce the sensitivity to singular value and noise data in simulation data, actual data and image segmentation, and clustering is more accurate and efficient.

Table of Content