Started in January,1974(Monthly)
Supervised and Sponsored by Chongqing Southwest Information Co., Ltd.
ISSN 1002-137X
CN 50-1075/TP
CODEN JKIEBK
Editors
Current Issue
Volume 45 Issue 9, 20 September 2018
  
Surveys
Analysis and Investigation of Research Frontiers in International Field of Artificial Intelligence in 2017
YAO Yan-ling
Computer Science. 2018, 45 (9): 1-10.  doi:10.11896/j.issn.1002-137X.2018.09.001
Abstract PDF(6387KB) ( 727 )   
References | Related Articles | Metrics
The literature co-citation analysis could provide a more objective and comprehensive perspective for the ana-lysis and investigation of the research frontiers in the target field.This paper analyzed 131 ESI highly cited papers in the international field of artificial intelligence in 2017 by literature co-citation analysis,and investigated 12 research frontiers and 2 key research frontiers in this field in 2017.Through further research on the core papers in the research frontiers,it is found that many Chinese scholars have been the backbones and play important roles.Comparatively speaking,in the two key research frontiers on deep learning,China still lacks the scholars producing high-quality core papers,and it needs further efforts of Chinese scholars.
Research Progress of Object Detection Technology Based on Convolutional Neural Network in Deep Learning
WANG Hui-ling, QI Xiao-long, WU Gang-shan
Computer Science. 2018, 45 (9): 11-19.  doi:10.11896/j.issn.1002-137X.2018.09.002
Abstract PDF(2074KB) ( 2243 )   
References | Related Articles | Metrics
Object detection is a hot topic in the field of computer vision.In recent years,convolutional neural network in deep learning has performed prominently in object detection tasks.This paper surveyed the research progress of deep learning in object detection.Firstly,two methods and commonly datasets of object detection were introduced and the advantages of deep learning based on object detection tasks were analyzed.Secondly,according to the development process of the object detection method based on deep learning,the classical convolutional neural network model used in this method was introduced,and the characteristics of each network model were analyzed.Then the aspects of the ability to acquire features,the speed of detection,and theused key technologies were analyzed and summarized.Finally,according to the difficulties and challenges existing in the object detection method based on deep learning and the future development trend,the thinking and outlook were made.
Survey on Recompression Detection for Digital Images
WANG Zhi-feng, ZHU Lin, ZENG Chun-yan, MIN Qiu-sha, XIA Dan
Computer Science. 2018, 45 (9): 20-29.  doi:10.11896/j.issn.1002-137X.2018.09.003
Abstract PDF(1876KB) ( 1393 )   
References | Related Articles | Metrics
With the wide application of digital image processing technology,the softwares of digital image processing have brought more convienience in our work and daily life,but a series of social problems caused by malicious tampered images also need to be solved,so the digital image forensics technology,which can judge the authenticity and integrity of the image,is particularly important.Since tampered images are always accompanied with the process of recompression,recompression detection can provide strong supporting evidence for digital image forensics.This paper systematically analyzed the current research of recompression detection,proposed a general framework for recompression detection,and elaborated the history detection of lossless images compression,double compression detection of loss images,multiple compression detection of lossless images,and recompression detection of other formats.This paper also analyzed and evaluated the performance of the existing algorithms,and then summarized the application of image compression detection.Finally,this paper analyzed the existing problems of recompression detection,andprospected the future development directions.
Review of Research on Image Complexity
ZHOU Bing, LIU Yu-xia, YANG Xin-xin, LIU Yang
Computer Science. 2018, 45 (9): 30-37.  doi:10.11896/j.issn.1002-137X.2018.09.004
Abstract PDF(1385KB) ( 3141 )   
References | Related Articles | Metrics
Image complexity has been not only widely studied in the field of computer science,but also extended to the field of medicine,cognitive psychology,etc.In the above application fields,different scholars have proposed different defi-nitions of complexity related to image,such as image complexity,visual complexity,scene complexity.And image complexity includes color complexity,shape complexity,and texture complexity.This paper summarized these applications,definitions and methods,and proposed a future research idea based on the definition of complexity in constitution theory to study the complexity from the point of image composition.This definition of the complexity only depends on the elements and contents of the image,and does not be related to the specific image processing algorithm.Meanwhile,this definition of complexity is consistent with the subjective perception and understanding of the image complexity in common sense.The elements of image can be classified to color,shape and texture feature.According to constitution theory,this paper proposed the definitions of the general sets for these three features.The further study is to calculate the image complexity and to verify its correlation with the subjective perception through experiments and correlation analysis
Application of Data Science in Management Science Study:State-of-the-art in Domestic
LI Zhi-guo, ZHONG Jiang
Computer Science. 2018, 45 (9): 38-45.  doi:10.11896/j.issn.1002-137X.2018.09.005
Abstract PDF(1405KB) ( 1575 )   
References | Related Articles | Metrics
As a new discipline,data science which is based on the era of big data isapplied to other disciplines including management.Firstly,the connections and differences between the scientific research paradigm based on data science and the classical paradigm of management research were proposed.Secondly,the relative literatures and citation literatures in the major Management Journal of A-class identified by the National Natural Science Foundation of China were identified,and the current management research hotspots like data-driven based public management,network behavior mana-gement based on complex network simulation and innovation management based on multisource data fusion were classified and summarized.And then,the characteristics of data science used in current management field were summarized.Finally,the trend of the application of data science in management science research was proposed,that is paradigm fusion,big data utilization,scene fusion and expert cooperation.
NASAC 2017
Linux Container Cluster Networking Approach for Multiple Tenants
ZHU Yu-jian, MA Jun-ming, AN Bo, CAO Dong-gang
Computer Science. 2018, 45 (9): 46-51.  doi:10.11896/j.issn.1002-137X.2018.09.006
Abstract PDF(2649KB) ( 963 )   
References | Related Articles | Metrics
At present,more and more cloud platforms begin to use Linux container cluster to provide runtime environment for cloud services.But how to build a stable and high-performance network for a user’s Linux container cluster under multi-tenant circumstance is an important technical problem.A networking approach of Linux container cluster for multiple tenants was proposed in this paper.Compared with that of Kubernetes,the proposed approach simplifies the network architecture and introduces network isolation.The network can meet the needs under multi-tenant circumstance.This paper described the design of the approach with a small and large scale of clusters and users and explained the implementation of it in a virtual cloud operating system Docklet.The source codes are open source on GitHub.Besides,evaluation results show that the performance of container network of the proposed approach is close to the original network.The TCP export downlink bandwidth is different from the original one within 0.4% and the TCP internal bandwidth gets about 3.39% loss.The batch job and long service applications are also well supported by the approach.
Evidence Model Oriented to Airborne Software Airworthiness Review of Software Planning Stage
YUAN Wei, WU Ji, LIU Chao, YANG Hai-yan
Computer Science. 2018, 45 (9): 52-59.  doi:10.11896/j.issn.1002-137X.2018.09.007
Abstract PDF(4547KB) ( 2486 )   
References | Related Articles | Metrics
Airworthiness certification is essential for airborne software.The 178C standard is an important safety certification standard for airworthiness.Order 8110.49 guideline sets out the method of airworthiness certification,but there is no research on the 178C target compliance evidence of the software planning phase involved in the review.Based on the objectives of the software planning phase and the feature description of the DO-178C standard,three models were proposed:standard evidence model,project-artifact model and project-related evidence model.And evidence information checklist isgenerated by converting the project-related evidence model to the project-related evidence data modelto determine the source of the evidence information.The review method for establishing the evidence model provides gui-dance for the collection of evidence for the review of the software planning phase,reducing the reliance on the auditor’sreview process and improving the efficiency of the review.And the availability and validity of the proposed evidence model were llustrated by an Airborne-Flight-Display software.
Web Based Lightweight Tool for Big Data Processing and Visualization
LI Yan, MA Jun-ming, AN Bo, CAO Dong-gang
Computer Science. 2018, 45 (9): 60-64.  doi:10.11896/j.issn.1002-137X.2018.09.008
Abstract PDF(3529KB) ( 1186 )   
References | Related Articles | Metrics
Researchers in the daily study often use Excel,Spss and other tools to analyze and process the data to obtain the knowledge of relevant field.However,with the arrival of large data age,due to constraints of stand-alone performance,general data processing software cannot meet the needs of researchers for large data analysis and processing.Large data processing and visualization are inseparable from the distributed computing environment.Therefore,in order to complete the rapid processing and visualization of large data,researchers not only need to purchase and maintain a distributed cluster environment,but also need to be able to program in a distributed environment and master the corresponding front-end data visualization technology.It is very difficult and unnecessary for many non-computer science data analysis workers.In view of the above problems,this paper presented a Web-based lightweight large data processing and visualization tool.Using this tool,data analysis workers can easily open a large data file(GB level) in the browser,quickly locate the file,sort the contents of the file and visualize it through a simple click and drag.At last,a correspon-ding empirical study was carried out to prove the effiectiveness of this solution.
Naive Bayesian Decision TreeAlgorithm Combining SMOTE and Filter-Wrapper and It’s Application
XU Zhao-zhao, LI Ching-hwa, CHEN Tong-lin, LEE Shin-jye
Computer Science. 2018, 45 (9): 65-69.  doi:10.11896/j.issn.1002-137X.2018.09.009
Abstract PDF(2734KB) ( 851 )   
References | Related Articles | Metrics
How to efficiently and accurately dig out the medical data generated by the Internet-based wisdom medical system with “Industrial 4.0” is still a very serious problem.However,the medical data is often high-dimensional,unba-lanced and noisy,so this paper proposed a new data processing method combining SMOTE method with Filter-Wrapper feature selection algorithm to support clinical decision-making.In particular,the proposed method not only overcomes the situation of bad prediction result of the independent assumptions in the practical attribute application of Naive Bayesian,but also avoids over-fitting problem caused by constructing the model of C4.5 decision tree.What’s more,when the proposed algorithm is applied to ECG clinical decision-making,good results can be obtained.
Study of Partition Mechanism for seL4 on Multi-core Platform
DING Gui-qiang, WANG Lei, WANG Lu-ming, KANG Qiao
Computer Science. 2018, 45 (9): 70-74.  doi:10.11896/j.issn.1002-137X.2018.09.010
Abstract PDF(1511KB) ( 1012 )   
References | Related Articles | Metrics
In avionics and other embedded fields,the multi-core era has come.How to make full use of multiple cores has become the research hotspot in field of the system.At the same time,as the system integration is getting higher and higher,a hardware platform may need to run tasks with different security levels,which requires the operating system to provide isolation and protection for the application.In order to solve the two problems mentioned above,this paper added multi-core and partition isolation support on the basis of seL4 with a single core,and then put forward multi-core seL4 and partition mechanism to achieve a partition mechanism with multi-core seL4 running on the qemu simulator.The implementation of the partition mechanism is in accordance with the semantics of the ARINC653 standard.
Modeling in Multiple Views and Industrial Case Study of Automatic Test for Hardware System
MENG Han, WU Ji, HU Jing-hui, LIU Chao, YANG Hai-yan, SUN Xin-ying
Computer Science. 2018, 45 (9): 75-80.  doi:10.11896/j.issn.1002-137X.2018.09.011
Abstract PDF(2107KB) ( 742 )   
References | Related Articles | Metrics
The development of ATE(Automatic Test Equipment) of hardwore system is a tedious task.It requires developers to understand various information about external ports,signals,test procedures and signal inspections of mea-sured equipment from specialists,thus comfirming the demand of development.In this process,the most complicated issue for developers is the lack of a normative model,which can describe the test information from various collaborators.This issue results in several problems of verboseness of test documents,difficulty of understanding and excess of errors.This paper proposed anATE domain-oriented multi-angle views modeling method.This model can normalize the test information of ATE and support the consistency checking.In the meantime,this paper gave an industrial case to demonstrate the effectiveness of the model.
Framework Assisting Storm Application Development Driven by Data Requirements
ZHOU Wen, SHI Xue-fei, WU Yi-jian, ZHAO Wen-yun
Computer Science. 2018, 45 (9): 81-88.  doi:10.11896/j.issn.1002-137X.2018.09.012
Abstract PDF(2366KB) ( 607 )   
References | Related Articles | Metrics
Storm,a widely used stream calculation framework,supports high efficient real-time calculation for stream data.In the development of Storm applications,developers have to write modules for various stream data requirements,causing repetitive work and difficulties in adapting to changes in data requirements.How to develop Storm applications and configure corresponding environment rapidly based on data requirements such as stream data format and calculations is an important research question for improving the efficiency of stream-oriented application development.An approach for describing stream data requirements was proposed in this paper.A framework assisting Storm application development was designed and implemented for business people to describe domain-specific data requirements and gene-rate Storm applications automatically.Experiments show that the framework is able to help non-developers configure and deploy common Storm-based stream calculation applications.The framework is adaptive to common requirements in real-time stream data calculations.
Study on Technical Debt Caused by Requirement Change
ZHANG Yun-jie, ZHANG Xuan, DING Hao, WANG Xu
Computer Science. 2018, 45 (9): 89-93.  doi:10.11896/j.issn.1002-137X.2018.09.013
Abstract PDF(2700KB) ( 721 )   
References | Related Articles | Metrics
In software life cycle,the requirement changes constantly,and the requirement decision often depends on developer’s preference and balance,and lacks a systematic and definite management method.Adefinition of technical debt caused by requirement change was proposed for the technical debt caused by the constant requirement change in software life cycle.Through the definition,detection,quantification and sorting of requirement change technical debt,technical support was provided for the realization sequence and realization way of requirement change.Finally,experiments were conducted to verify the feasibility of concept and technology for technical debt caused by requirement change.
Signal Model of Automatic Testing Technology for Airborne Equipment
HU Jing-hui, WU Ji, MENG Han, LIU Chao, YANG Hai-yan
Computer Science. 2018, 45 (9): 94-98.  doi:10.11896/j.issn.1002-137X.2018.09.014
Abstract PDF(1820KB) ( 573 )   
References | Related Articles | Metrics
In the airborne equipment testing,exact definition and detection of various test signals is a key factor to ensure safety flight.This paper presented a signal model for the test requirements description of automatic test airborne equipment.In the research,the signals involved in the test requirements of the airborne equipment were categorized,the signals transmitted between the test equipment and the equipment under test were regarded as a set of action including parameters,and a signal model for automatic test airborne equipment requirements was established.The model aims to address the problem of the signal specification description in the airborne equipment automatic test equipment area of current industry.This paper illustrated the characteristics and feasibility of the signal model through a typical case study.
Rule-driven DFS Testing Technology for Android Application
YE Jia, GE Hong-jun, CAO Chun, ZHU Jin, ZHANG Ying
Computer Science. 2018, 45 (9): 99-103.  doi:10.11896/j.issn.1002-137X.2018.09.015
Abstract PDF(2476KB) ( 890 )   
References | Related Articles | Metrics
Automated GUI testing is the important part of Android application research field.Several technologies for automated Android GUI testing have attracted wide attention.Testing technology based on DFS exploration has been extensively used among them.However,the existing DFS testing technology is still inefficient and has low testing co-verage.This paper proposed an improved approach by driving the DFS automated exploration with external predefined rulesto improve the efficiency and coverage.A testing tool called RDTA based on the proposed approach was implemented and the performance of RDTA was evaluated by comparing to Monkey and original DFS without rules.The result verifies the effectiveness of the approach.
Design and Implementation of Distributed Full-text Search Framework Based on Spark SQL
CUI Guang-fan, XU Li-jie, LIU Jie, YE Dan, ZHONG Hua
Computer Science. 2018, 45 (9): 104-112.  doi:10.11896/j.issn.1002-137X.2018.09.016
Abstract PDF(4086KB) ( 726 )   
References | Related Articles | Metrics
With the development of information technology,big data has generated great value in various fields.Huge data storage and rapid analysis have become new challenges.The traditional relational database is difficult to meet the needs of big data storage and analysis because of its shortcomings in terms of performance,scalability and high cost.Spark SQL is a data analysis tool based on Spark,which is a big data processing framework.Spark SQL currently supports the TPC-DS benchmark and has become an alternative solution to the traditional data warehouse under the background of big data.Full-text search,as a kind of effective method of text search,can be used in combination with general query operation to provide richer queries and analysis operations.Spark SQL doesn’t support full-text search now.In order to meet the needs of traditional business migration and existing business,this paper proposed a Spark SQL distributed text retrieval framework,covering the design and implementation of 4 modules including SQL grammar,SQL translation framework,full-text search parallelism and search optimization.The results of experiment show that,under the two search optimization strategies,index construction time and query time of this framework are reduced to 0.6%/0.5% and 1%/10% respectively compared with the traditional database,and index storage volume is reduced to 55.0%.
Big Data Driven Analysis of Knowledge Exchange Network in Developer Community
DA Yi-fei, LIU Xu-dong, SUN Hai-long
Computer Science. 2018, 45 (9): 113-118.  doi:10.11896/j.issn.1002-137X.2018.09.017
Abstract PDF(1695KB) ( 684 )   
References | Related Articles | Metrics
Developer community generally has various sections such as blog,QA,bbs,etc.These sections together constitute a platform where users contribute and communicate software development knowledge.This paper concentrated and analyzed the big data accumulated in CSDN and constructed knowledge exchange networks.Analysis of the networks based on complex network method indicates the small-world effect and scale-free property of multi-section knowledge exchange network.Further study shows that there exist relatively more knowledge distributors in multi-section users and they play a relative more important role in the knowledge exchange network.
Research and Implementation of Collaborated Modeling Approach for Problem-oriented Software Development
ZHANG Xiao, LI Zhi, ZHAO Zi-yan, FU Chang-lan, LI Wei-dong, YU Yue-kun, WANG Chao
Computer Science. 2018, 45 (9): 119-122.  doi:10.11896/j.issn.1002-137X.2018.09.018
Abstract PDF(3259KB) ( 721 )   
References | Related Articles | Metrics
Software modeling tools are essential for assisting requirements engineers in system analysis during the requirements and design phase.At present,few existing requirements modeling tools can be run across different platforms,support online multi-user collaborations,and verify the correctness and completeness of requirements models.As the problem frames(PF) approach attracts much attention in the requirements engineering community,a computer-aided PF modeling tool,which provides good user experience and is compatible with multiple platforms,was developed in this paper.This work solves two difficult problems,i.e.,automatic verification of the correctness and completeness of problem diagrams and mechanized decomposition of complex problem diagrams.Therefore,an online platform for requirements modeling,sharing and verification was established to support multi-user logins,deploy databases in the clouds and facilitate multi-user collaborations.
Network & Communication
Cluster-based Radio Resource Allocation Mechanism in D2D Networks
LI Fang-wei HUANG Xu ZHANG Hai-bo LIU Kai-jian HE Xiao-fan
Computer Science. 2018, 45 (9): 123-128.  doi:10.11896/j.issn.1002-137X.2018.09.019
Abstract PDF(4578KB) ( 627 )   
References | Related Articles | Metrics
A cluster-based resource allocation mechanism was proposed for the problem of the limited battery life of the user equipment.The mechanism includs three parts:D2D users are put in disjoint clusters by graph theory,and then the auction algorithm is used to allocate the channel for D2D clusters according to the clustering results.In the end,the power allocation is performed by using non-cooperative game theory model.Simulation results show that the proposed mechanism effectively extends battery life,improves the energy efficiency of the users,and satisfies the requirements of the user transmission rate,thus ensuring the users to obtain higher QoS.
HMM Cooperative Spectrum Prediction Algorithm Based on Density Clustering
WU Jian-wei, LI Yan-ling, ZHANG Hui, ZANG Han-lin
Computer Science. 2018, 45 (9): 129-134.  doi:10.11896/j.issn.1002-137X.2018.09.020
Abstract PDF(3756KB) ( 714 )   
References | Related Articles | Metrics
Aiming at the problems of long time delay and low prediction accuracy in traditional hidden Markov spectrum prediction,this paper proposed an HMM cooperative spectrum prediction algorithm based on density-based spatial clustering of applications with noise (DBSCAN).BSCAN algorithm is used to cluster the frequency domain channels with strong correlation and predict the channel state in units of clusters,and the prediction delay is reduced by reducing the number of predicted times.At the same time,the method of multiple sub-users cooperative prediction is used in the time domain,and the forecast uncertainty is reduced by fusing the initial prediction results of each subordinate user.Simulation results show thatthe proposed algorithm has shorter spectral delay and higher accuracy compared with the traditional HMM-based local spectrum prediction algorithm and HMM-based packet fusion prediction algorithm.
Heterogeneous Chain Dominating Set Algorithm in Wireless Ad Hoc Networks
HAN Bing-qing, CHEN Yi-fei
Computer Science. 2018, 45 (9): 135-140.  doi:10.11896/j.issn.1002-137X.2018.09.021
Abstract PDF(2735KB) ( 668 )   
References | Related Articles | Metrics
Firstly,the heterogeneous disk model HDG of Ad Hoc network was presented,and different forms of HDG model were analyzed.Secondly,a new two-way chain table structure was designed.Based on the chain structure,a connected dominating set algorithm C-LDS was proposed,which manages dominating set through two-way chain table structure,and improves the time efficiency of inserting,deleting and modify the dominating node through the node refe-rence way.Thus,the connected dominating set of Ad Hoc network is optimized.Finally,the C-LDS algorithm was compared with other dominating set algorithms.The simulation results show that C-LDS algorithm generates minimal CDS size in both uniform and random scenarios,and can achieve highest packet delivery ratio in mobile network scenarios,which showsgood heterogeneous connectivity and improves the efficiency yeneration of dominating set node.
Simulation Research on Improved Decoding Algorithm Based on Non-binary LDPC for 5G
MENG Jia-hui, ZHAO Dan-feng, TIAN Hai
Computer Science. 2018, 45 (9): 141-145.  doi:10.11896/j.issn.1002-137X.2018.09.022
Abstract PDF(2004KB) ( 979 )   
References | Related Articles | Metrics
According to the characteristics of high speed,low delay,high capacity data transmission and various scenes in 5G mobile communication,this paper proposed a low complexity decoding algorithm based on Non-binary LDPC codes for 5G,which is called mixed Log-FFT-BP decoding algorithm.The algorithm directly computes the logarithm of the probability information and avoids the operation of computing the log likelihood ratio.In the process ofupdating the check nodes,the algorithm uses the logarithm of the intermediate variable and the inverse Fourier transform of the check node information to reduce the decoding complexity.The simulation was conducted for 5G channel coding to support a wider range of code block lengths and more code rates.The simulation results show thatthe performance of the improved mixed log-FFT-BP algorithm has little difference (about 0.1~0.2dB) compared with the traditional decoding algorithm,which reduces the decoding complexity and is more conducive to the realization of hardware platform
Research on Stochastic Resonance Characteristics of Cascaded Three-steady-state and Its Application
ZHANG Gang, GAO Jun-peng, LI Hong-wei
Computer Science. 2018, 45 (9): 146-151.  doi:10.11896/j.issn.1002-137X.2018.09.023
Abstract PDF(2042KB) ( 792 )   
References | Related Articles | Metrics
In order to solve the problem of weak signal detection difficulties in strong noise environment,using SNR gain and the spectral height of characteristic frequency as the measurement indexes,this paper studied the cascaded tri-stable stochastic resonance system and analyzed its characteristics.The simulation results show that the cascaded tri-stable stochastic resonance system can achieve better output than single-stage tri-stable resonance system through tu-ning the parameters.In addition,in order to solve the problem that the weak signal in the actual gear fault diagnosis is difficult to extract,this paper proposed a gear fault diagnosis method by using cascaded tri-stable stochastic resonance system.The results show that this method can effectively extract the weak characteristics of gear fault,and realize the early gear fault diagnosis.Therefore,it has a wide range of engineering application prospects.
New Q Value Anti-collision Algorithm Based on Label Grouping
YANG Zi-wei, ZHENG Jia-li, YUE Shi-bin, YUAN Yuan, SHI Jing
Computer Science. 2018, 45 (9): 152-155.  doi:10.11896/j.issn.1002-137X.2018.09.024
Abstract PDF(3552KB) ( 691 )   
References | Related Articles | Metrics
RFID technology is a key technology of the Internet of things.In order to solve the problem of anti-collisionin RFID technology with a large number of tag data,a new Q value algorithm based on EPC-C1G2 was proposed in this paper.In this algorithm,the method of Q is improved effectively.Otherwise,combined with the label grouping algorithm,the proposed method achieves higher system efficiency in a large number of tag data.The simulation results show that the proposed algorithm can not only reduce the slot number,improve the time slot utilization,but also maintain a good system throughput rate.
Multi-channel Power Control Mechanism Based on Hidden Markov in Cognitive Network
ZHU Jiang, MA Xiao, YIN Yao-hu
Computer Science. 2018, 45 (9): 156-160.  doi:10.11896/j.issn.1002-137X.2018.09.025
Abstract PDF(1566KB) ( 544 )   
References | Related Articles | Metrics
In the distributed multi-channel access cognitive radio network,to deal with the resource allocationconflict issue caused by asgmmetric environmental information of users,according to the correlations of channel detected results of unlicensed users,a multi-channel game-theoretic power control mechanism based on Hidden Markov model was proposed.The mechanism selectes reasonable price function to effectively suppress the selfish behavior of unauthorized users,and realizes the spectrum sharing between unlicensed users and makes them estimate whether other users on the channel would take part in the game,thus obtaining more accurate information about the game and choosing a better transmission power.The simulation results show that the system can achieve higher efficient capacity and ensure that more users meet the speed requirements.
Compressive Sensing Multi-target Localization Algorithm Based on Data Fusion
YANG Si-xing, GUO Yan, LI Ning, SUN Bao-ming, QIAN Peng
Computer Science. 2018, 45 (9): 161-165.  doi:10.11896/j.issn.1002-137X.2018.09.026
Abstract PDF(13213KB) ( 786 )   
References | Related Articles | Metrics
This paper proposd a new compressive sensing localization algorithm based on data fusion,which can utilize all kinds of the localization data at the same time.The proposed theory is based on the sparsity of the target number and compressive sensing theory,and it can greatly reduce the quantity of sampling compared with the traditional localization algorithms.The new algorithm consists of data pre-processing and data fusion based localization.At the first step,different kinds of measurements are transferred into the form which has the same level.Then the technique of multiple measurement vectors is used to recover the target vector.Compared with other algorithms,the proposed algorithm holds better performance in localization accuracy and robustness.
Double Thresholds DMM Cooperative Spectrum Sensing Algorithm Based on Credibility
GAO Peng, LIU Yun-jiang, GAO Wei-ting, LI Man, CHEN Juan
Computer Science. 2018, 45 (9): 166-170.  doi:10.11896/j.issn.1002-137X.2018.09.027
Abstract PDF(3476KB) ( 565 )   
References | Related Articles | Metrics
Because the eigenvalue-based double thresholds spectrum sensing algorithms overlook the reliability differencebetween second users(SU) and the highexpense of fusion decisions,a double thresholds DMM cooperative spectrum sensing algorithm based on credibility(DT-CDMM) was proposed to improve the sensing performance.Based on the difference between maximum and minimum eigenvalue(DMM) algorithm,a double thresholds DMM spectrum sensing algorithm based on limiting eigenvalue distribution is used as SU’s local sensing,a triggered mechanism combined with soft and hard decisions is established to cut the system expenses,the final decision is obtained via the weighting of SU’ssensing performance and local credibility,and a self-adaption compensation for hard decisions is applied.Theory analysis and simulations show that the DT-CDMMimproves the probability of multi-user collaborative detection compared with double eigenvalue thresholds algorithms and double thresholds energy detection algorithm when noise is undefined.
Information Security
Mining RTSP Protocol Vulnerabilities Based on Traversal of Protocol State Graph
LI Jia-li, CHEN Yong-le, LI Zhi, SUN Li-min
Computer Science. 2018, 45 (9): 171-176.  doi:10.11896/j.issn.1002-137X.2018.09.028
Abstract PDF(1581KB) ( 1263 )   
References | Related Articles | Metrics
Currently,many video surveillance equipments like cameras,DVRs,and NVRs support RTSP protocol,and the number of buffer overflow vulnerabilities caused by the RTSP protocol is large and harmful.Therefore,the research on the RTSP protocol has both application value and theoretical significance.The number of test cases generated by directly using the fuzzy test framework is huge,and the test process takes a long time.Aiming at the above problems,this paper took the RTSP protocol of video surveillance equipment as the research object,and proposed a method which removes duplicate sample set of the protocol basic block,uses the constraint relationship and state transition between protocol states to construct protocol state diagram,and dose deep traversal based on protocol state diagram.This method reduces the generation of test cases and improves the effectiveness of generation.When the RTSP protocol is tested by fuzzing method,the method of sending a TCP probe packet is used to determine whether the test target is abnormal.The redundant part of the recorded abnormal test case is removed,which facilitates subsequent playback and reduces the time, thereby improving the efficiency of vulnerability mining.
New Cross-domain Authentication Model for Information Services Entity
XIE Yan-rong, MA Wen-ping, LUO Wei
Computer Science. 2018, 45 (9): 177-182.  doi:10.11896/j.issn.1002-137X.2018.09.029
Abstract PDF(1426KB) ( 878 )   
References | Related Articles | Metrics
To solve the problem that the identity of information services entity(ISE) cannot be revoked immediately in the cross-domain authentication system,a revocable identity-based signature scheme was proposed.Based on the SM9 signature algorithm,a security mediator(SEM) was introduced to keep a part of the private key of the ISE.By terminating the SEM to send the token to ISE to revoke its signature capability,the identity of ISE can be revoked immediately.Based on this scheme,a new cross-domain authentication model for ISE was proposed by taking the combining advantages of certificate-based public key infrastructure(PKI) and identity-based cryptography(IBC).The proposed model is not only flexible and efficient,but also suitable for constructing large-scale application environment of ISE.Meanwhile,a cross-domain authentication protocol was designed to realize the mutual authentication with key agreement between cross-domain entities.Analysis shows that the proposed protocol has high security and low communication and computation cost.
Security Analysis of Heterogeneous Redundant Systems
WANG Wei, YANG Ben-chao, LI Guang-song, SI Xue-ming
Computer Science. 2018, 45 (9): 183-186.  doi:10.11896/j.issn.1002-137X.2018.09.030
Abstract PDF(2272KB) ( 1360 )   
References | Related Articles | Metrics
With the development and popularization of Internet technology,vulnerability and backdoor problems have become the main factor of network security problems.The redundancy technology can solve the reliability problem of system.Inspired by the idea of the mimicry defense,this paper analyzed the effectiveness of the heterogeneous redundant technology against the security defense based on the vulnerability and backdoor network attack.On some assumptions,this paper established a security model of heterogeneous redundant system based on Markov process.System security was characterized by the success rate of system attack,and the expression of success rate of system attack was given.At last,triple-redundant heterogeneous system was solved and analyed.The experimental results are in accordance with the intuitive expectations.
Attribute Revocable Access Control Scheme for Brain-Computer Interface Technology
WANG Jing, SI Shu-jian
Computer Science. 2018, 45 (9): 187-194.  doi:10.11896/j.issn.1002-137X.2018.09.031
Abstract PDF(1749KB) ( 609 )   
References | Related Articles | Metrics
Though brain-computer interface(BCI) technology has wide application in the field of rehabilitation medicine,the general neglect of private sensitive data protection usually leads to serious security issues.In this paper,a secure and efficient attribute-based access control scheme was proposed for the privacy protection in BCI applications.The new scheme uses the version number tag and proxy re-encryption technology to realize the attribute revocation,which makes the access strategy more flexible.The experimental results show the scheme’s capability enhancing the computational efficiency and reducing the computational complexity,as well as its effectiveness in the privacy protection of the BCI system.
Reversible Visible Watermarking Algorithm for Medical Image Based on Support Vector Regression
WANG Nan, LI Zhi, CHENG Xin-yu, CHEN Yi
Computer Science. 2018, 45 (9): 195-201.  doi:10.11896/j.issn.1002-137X.2018.09.032
Abstract PDF(8022KB) ( 698 )   
References | Related Articles | Metrics
With the development of medical imaging technology,medical image has become the main basis for doctors to diagnose patients’ condition.In order to provide more accurate diagnosis and optimal treatment plan for patients,medical image sharing and expert remote diagnosis have become important medical tools.Unprotected medical images are easily attacked or tampered maliciously during transmission,so this paper proposed a reversible visible watermarking algorithm for medical image based on support vector regyession to effectively protect the integrity of medical image information and limit the application of unauthorized users.Firstly,the algorithm embeds the visual watermark into the medi-cal image.Secondly,it uses the support vector regression to forecast the pixel value of the watermark image,and calculates the prediction error.Then,it applies the texture level to determine the appropriate brightness adjustment thre-shold,thus generating a global positioning map according to the threshold,and it adds the global positioning map to the original image to enhance the robustness of the algorithm.Finally,it generates a reversible visual watermark by using the relationship between the global positioning map and the prediction error to encrypt the medical image.The experimental results show that the reversible visible algorithm can be applied to not only traditional medical images,but also diffusion weighted images for the first time.The algorithm has good robustness and reversibility,and can effectively avoid pixel overflow problems,achieving the correct extraction of reversible watermark.There is no any difference between the recovered image and the original image.When the watermark information is unknown,it is difficult to remove visible reversible watermarking,and this feature can effectively protect the integrity of patient information and the authority of medical images.
Fault Tree Structure Matching Algorithm and Its Application
YUE Xin, DU Jun-wei, HU Qiang, WANG Yan-ping
Computer Science. 2018, 45 (9): 202-206.  doi:10.11896/j.issn.1002-137X.2018.09.033
Abstract PDF(2332KB) ( 807 )   
References | Related Articles | Metrics
A large number of fault trees have been designed and stored with the occurrence of numerous historical accident cases.Structure matching is an effective way to achieve accurate and comprehensive investigation of new accidents by using the existing fault trees with the limited time,manpower and cost.Based on the timing of event evolution and the structural features of causal reasoning,a fault tree structure matching algorithm was proposed.The hidden Markov model of fault tree is constructed and then the Viterbi algorithm is used to predict the optimal matching sequences.Compared with the node-based structure matching algorithm,this algorithm has significant improvement in the accuracy of matching and the detection of structural defects.
Software & Database Technology
New Spectrum-based Fault Localization Method Combining HittingSet and Genetic Algorithm
ZHOU Ming-quan, JIANG Guo-hua
Computer Science. 2018, 45 (9): 207-212.  doi:10.11896/j.issn.1002-137X.2018.09.034
Abstract PDF(1405KB) ( 606 )   
References | Related Articles | Metrics
Fault localization is an important research topic in the process of software development.However,the number of faults in the actual software cannot be determined in advance.The available single fault localization technique is not convenient to be used,and the available multi-fault localization technique is of low locating efficiency.This paper studied and improved the multi-fault localization technique GAMFL,and proposed a new spectrum-based fault localization methoid combining hitting set and genetic algorithm(GAHIT).In this method,the basic block for localization is defined and used to replace statements to localize faults,narrowing the search range.In the process of constructing initial population,the method of solving the hitting sets of execution path of failure test cases is presented to optimize the generation of initial population,and a new method for calculating fitness function is also presented to improve the total efficiency of the algorithm.According to the results of genetic algorithm,the fault detecton strategy is presented to improve the accuracy of localizing faults in the optimal group.The experiment results show that the proposed method is effective in solving the problem of localizing programs with unknown number of faults,and has good performance when localizing faults in both single fault programs and multi-fault programs.
Spatial Index of 3D Point Cloud Data Based on Spark
ZHAO Er-ping, MENG Xiao-feng
Computer Science. 2018, 45 (9): 213-219.  doi:10.11896/j.issn.1002-137X.2018.09.035
Abstract PDF(1486KB) ( 1076 )   
References | Related Articles | Metrics
Two level spatial index based on R tree was presented according to the problem that spark engine doesn’t support multi-dimensional spatial query,that is,the R subtree is created on each worker node,and these subtrees are used as children to create the R tree on the master node.Memory replacement granularity of LRU algorithm is coarse,and the result is not accurate enough.For this reason,the method of memory replacement based on data usage weight was proposed.The ratio of actual used amount of data and its total amount is used as replacement weight.The method stores the hot scene data in RDD form into memory and improves the query efficiency based on memory.According to the visual principle of far thick and near fine,the level of detail query was presented.The point cloud data that best represent the object characteristics are firstly transmitted or the simplified model data are only transmitted to the client,so as to solve the problem of insufficient network bandwidth and data loading delay.Experimental results show that the proposed method can effectively solve the problem of multi-dimensional spatial query on spark,and the query efficiency is improved obviously.
Vectorization Methods for Indirect Array Index
YAO Jin-yang, ZHAO Rong-cai, WANG Qi, LI Ying-ying
Computer Science. 2018, 45 (9): 220-223.  doi:10.11896/j.issn.1002-137X.2018.09.036
Abstract PDF(2346KB) ( 662 )   
References | Related Articles | Metrics
Indirect array index cannot bevectorized efficiently in the existing compiler.It makes the program which contains the indirect array index cannot take advantage of SIMD extension parts.It is a hot topic in research on procedure vectorization.In order to utilize the SIMD extension parts efficiently and excavate the vectorization potential in the program fully,a new vectorization method for indirect array index was proposed in this paper.The performance income method was provided so as to analyze the performance benefits for various indirect arrays index.The experimental results show that the vectorization method can significantly improve the efficiency of the execution of program.
XACML Policy Query Method Based on Attribute And/Or Matrix and Type Analysis
HAN Dao-jun, YUAN Wan-li, DUAN Xiao-yu, ZHANG Lei
Computer Science. 2018, 45 (9): 224-229.  doi:10.11896/j.issn.1002-137X.2018.09.037
Abstract PDF(1491KB) ( 582 )   
References | Related Articles | Metrics
The description and execution of access control policy is an important way of information resource protection,which affects system’s operational running.In view of the poor efficiency of evaluation,some researchers have proposed the policy evaluation methods based on attribute cache and reordering,which improve the efficiency of policy eva-luation,but they still fail to solve the problem that the policy evaluation needs to traverse all relevant rules.To focus on this problem,after the analysis about the characteristics of the XACML policy description,a XACML policy query method based on attribute and/or matrix and type analysis was proposed in this paper,which can reduce the number of matching during policy evaluation.This method modifies the processing of the existing Context Handler,and adds a preprocessing phase which will match access control rule.During the preprocessing phase,the discriminations are calculated for each rule attributes.The irrelative rules for current access control request can be filtered by the attribute and/or matrix and the discriminations.The proposed method can improve the efficiency of policy evaluation by matching the filtered rule set.Experimental results verify its efficiency.
Artificial Intelligence
Two Types of Dynamic Information Law Models and Their Applications in Information Camouflage and Risk Identification
REN Xue-fang, ZHANG Ling, SHI Kai-quan
Computer Science. 2018, 45 (9): 230-236.  doi:10.11896/j.issn.1002-137X.2018.09.038
Abstract PDF(1430KB) ( 454 )   
References | Related Articles | Metrics
Function P-set is the function form of P-set.It is an information law model with dynamic characteristics and law or function characteristic obtained by improving P-set.In function P-set,the function’s attribute satisfies conjunctive normal form in mathematical logic.Function inverse P-set is the dual model of function P-set,in which the function’s attribute satisfies disjunctive normal form in mathematical logic.Here function P-set is defined as a type of dynamic information law model,while function inverse P-set is defined as another type of dynamic information law model.Based on the structures,characteristics and attribute normal forms of function P-sets and function inverse P-sets,a simple application of function P-set in information image splicing and camouflage and a simple application of function inverse P-set in profit risk estimation-identification were researched.Function P-sets and function inverse P-sets are the new theories and new models for the application research of dynamic information law.
Lawyer Evaluation Method Based on Network Response
YANG Kai-ping, LI Ming-qi, QIN Si-yi
Computer Science. 2018, 45 (9): 237-242.  doi:10.11896/j.issn.1002-137X.2018.09.039
Abstract PDF(1597KB) ( 679 )   
References | Related Articles | Metrics
With the development of society and the Internet,the citizen’s legal consciousness is gradually raised.Hence,the traditional business process and developmodels for lawyers are unsuitable for customers as well as the industry.Based on the existing response standards of professional lawyer consultation,the criteria for judging the response quality was proposed in this paper.Moreover,the response texts were quantitatively described from 5 aspects.Based on the word2vec algorithm,the similarity between word vectors and corresponding words was obtained from the existing database of lawyer question and answer system.Furthermore,the similarity function of texts was proposed based on word similarity and text length.Consequently,the quality evaluation model of the response of lawyers was established.Simulations were given to verify the validity of the model.The results show that the proposed model works well in evaluating the response quality of lawyer after the quantitative analysis of the question and answer text of each lawyer in the database.
Approach of Stance Detection in Micro-blog Based on Transfer Learning and Multi-representation
ZHOU Yan-fang, ZHOU Gang, LU Zhong-lei
Computer Science. 2018, 45 (9): 243-247.  doi:10.11896/j.issn.1002-137X.2018.09.040
Abstract PDF(1636KB) ( 955 )   
References | Related Articles | Metrics
Stance detection aims to identify users’ opinion towards a particular target.Aiming at the problem that exi-sting methods are often difficult to overcome the lack of labeled data and the error caused by word segmentation of Chinese text,this paper presented a transfer learning method and a hybrid model of character-level and word-level features.Firstly,character-level and the word-level features are inputted to deep neural network and the outputs of both are concatenated to reproduce the missing semantic information caused by word segmentation.Then,a topic classification model(parent model) is trained with a large external micro-blog data to obtain the effective sentence feature representation.Next,some of parent model’s parametersare used to initialize stance detection model and the knowledge transferred from auxiliary data can be used to enhance semantic representation ability of sentences.Finally,the labeled data are used to fine tune the child model andtrain classifiers.Experiment on NLPCC-2016 Task 4 proves that F1 value of proposed method achieves 72.2%,which is better than the best one of participating teams.The results show that this approach can improve the stance detection performance and alleviate the influence caused by word segmentation.
Sentiment Analysis of Hotline Data in Gas Industry
ZHU Hu-chao, YU Hui-qun, FAN Gui-sheng, DENG Cun-bin
Computer Science. 2018, 45 (9): 248-252.  doi:10.11896/j.issn.1002-137X.2018.09.041
Abstract PDF(1654KB) ( 543 )   
References | Related Articles | Metrics
Sentiment analysis of customer service hotline plays a decisive role in the development of enterprise core businesses,and can enhance customers’ loyalty.Traditional hotline emotional analysis methods use the ways of manual recording or random sampling,which not only consume manpower but also can’t guarantee accuracy,and the main problem is it cannot reflect customer’s emotion objectively,and ultimately affects the quality of service enterprises.Accor-ding to the background of the project and the existing offline audio files of Gas Company,hybrid algorithm of acoustic features and domain sentiment lexicon was proposed,which is used in the data analysis of customer service hotline and identifying customer sentiment(negative,non-negative).The experimental results show that the algorithm has an efficient recognition effect on the project practice,especially the combination of field of the sentiment lexicon.
Real-time Personalized Micro-blog Recommendation System
LIU Hui-ting, CHENG Lei, GUO Xiao-xue, ZHAO Peng
Computer Science. 2018, 45 (9): 253-259.  doi:10.11896/j.issn.1002-137X.2018.09.042
Abstract PDF(2287KB) ( 734 )   
References | Related Articles | Metrics
At present,many social networking services do not fully consider the personalized needs of users,and it is difficult to guarantee the real-time services because social networking services need to deal with massive amounts of data.A micro-blog recommendation model called RPMPS based on LDA topic model and KL divergence was proposed to respond to users’ personalized request in micro-blog recommendation in real time and improve the efficiency and quality of recommendation.RPMPS model not only uses the document-topic probability distribution matrix to get the similarity between the topic of user information and the topic of candidate micro-blog,but also obtains the similarity between the content of user information and the content of candidate micro-blog by utilizing the document-word to count the probability of the word frequency .At last,the real-time personalized micro-blog recommendation system based on RPMPS model is constructed,and micro-blog is filtered during the course of data processing to shorten the system response time.Experimental results on real-world datasets demonstrate that the system can meet the real-time personalized demands of users.
NKSMOTE Algorithm Based Classification Method for Imbalanced Dataset
WANG Li, CHEN Hong-mei
Computer Science. 2018, 45 (9): 260-265.  doi:10.11896/j.issn.1002-137X.2018.09.043
Abstract PDF(1408KB) ( 596 )   
References | Related Articles | Metrics
In SMOTE(Synthetic Minority Over-sampling TEchnique),only minority class samples nearest to neighbors are computed when samples are synthesized,causing the problem that the density of the minority class samples remains unchanged after oversampling.This paper proposed an improved NKSMOTE(New Kernel Synthetic Minority Over-Sampling Technique) algorithm to overcome the shortage of SMOTE.Firstly,a nonlinear mapping function is used to map samples to a high-dimensional kernel space,and then the K nearest neighbors of samples of minority class from the whole samples are computed.In addition,different over-sampling rates are set on different minority samples to change the imbalanced multiplying power according to the influencecaused by the distribution of minority class samples on the classification performance of algorithm.In the experiments,some classical oversampling methods were compared with the proposed oversampling method,and Decision Tree(DT),error BackPropagation(BP) and Random Forest(RF) were chosen as base classifier.Experimental results on UCI data sets show better classification performance of NKSMOTE algorithm.
Multi-feature Fusion for Short Text Similarity Calculation Based on LDA
ZHANG Xiao-chuan, YU Lin-feng, ZHANG Yi-hao
Computer Science. 2018, 45 (9): 266-270.  doi:10.11896/j.issn.1002-137X.2018.09.044
Abstract PDF(1995KB) ( 1015 )   
References | Related Articles | Metrics
In recent years,latent dirichlet allocation(LDA)topic model provides a new idea for short text similarity calculation by mining the latent semantic themes of text.In view of the sparse features of short text,because the application of LDA theme model may easily lead to inaccurate results of similarity computation,this paper presented a calculation method based on LDA model combining similarity topics factor ST and co-occurrence words factor CW to establish union similarity model.In the protocol of different ST intervals,CW generates constraint or supplementary conditions to ST,and obtains higher accuracy of text similarity.A text clustering experiment was used to verify the method.The experimental results show that the proposed method gains a certain improvement of F measure value
Forecasting of Medium and Long Term Precipitation Based on Hybrid Model
LI Dong, XUE Hui-feng
Computer Science. 2018, 45 (9): 271-278.  doi:10.11896/j.issn.1002-137X.2018.09.045
Abstract PDF(3543KB) ( 776 )   
References | Related Articles | Metrics
Accurate estimation of precipitation is an important issue in water resources engineering,management and planning.In order to improve the accuracy of medium and long term precipitation forecasting,a hybrid forecasting model based on modified ensemble empirical mode decomposition,least squares method,kernel extreme learning machine and modified fruit fly optimization algorithm was presented.By using modified ensemble empirical mode decomposition,non-stationary precipitation time series is decomposed into many terms.Then the decomposed terms are predicted by the least square method or the kernel extreme learning machine according to its characteristics.Because the kernel extreme learning machine has some characteristics of parameter sensibility,the modified fruit fly optimization algorithm is used to search the optimal parameters to improve the forecasting accuracy.Finally,forecast results of each decomposed term are added together to obtain the final forecasting results.The method was tested by using annual precipitation data from seven cities in China’s Guangdong province between 1951 and 2015.Results show that compared with the auto-regressive moving average model and kernel extreme learning machine model,the mixed model has higher prediction accuracy.
Graphics, Image & Pattern Recognition
Automatic Color Image Segmentation Algorithm Based on Random Region Merging
GU Wei-dong, LI Bing
Computer Science. 2018, 45 (9): 279-282.  doi:10.11896/j.issn.1002-137X.2018.09.046
Abstract PDF(4918KB) ( 655 )   
References | Related Articles | Metrics
In order to solve the problem of low segmentation accuracy for color image,a new algorithm for automatically segmenting color image with multi-scale spatial constraints was proposed.Based on the improved random region merging method,this algorithm firstly implements the bilateral decomposition and performs the over segmentation based on the multi-channel information and the multi-scale gradient.Then,in the CIE L*a*b* color space,a normalized color histogram is adopted to represent each sub-region.Finally,a stochastic region merging strategy with spatial constraints is constructed on the region adjacency graph to construct a segmentation graph for each scale.The experimental results in BSDS image database demonstrate that the proposed method has better segmentation performance than existing algorithms.
Multi-target Localization Method Based on FAsT-Match Algorithm
CHEN Jun, ZHENG Hong-yuan
Computer Science. 2018, 45 (9): 283-287.  doi:10.11896/j.issn.1002-137X.2018.09.047
Abstract PDF(4968KB) ( 568 )   
References | Related Articles | Metrics
FAsT-Match algorithm can realize the fast and accurate positioning ofthe template in a continuous image in the case of two-dimensional affine transformation.FAsT-Match algorithm is insensitive to light changes and has strong robustness.However,for images with multiple targets,only an approximate global optimal solution can be located.Therefore,the FAsT-Match algorithm was improved in this paper,and the target region obtained by fuzzy c-means clustering was used as the new target image,and then the original FAst-Match algorithm was used to locate the new target position and return to the original target image.This method can make up the shortcomings of only locating single target in FAsT-Match algorithm,reduce the hardware cost and locate target fast and accurately when it is applied in the wireless laser shooting system.The experimental results show that the method is effective and can meet the needs of positioning multiple targets,and has certain practical value.
Classification of Hyperspectral Remote Sensing Imagery Based on Second Order Moment Sparse Coding
XU Jia-qing, WAN Wen, LV Qi
Computer Science. 2018, 45 (9): 288-293.  doi:10.11896/j.issn.1002-137X.2018.09.048
Abstract PDF(5747KB) ( 738 )   
References | Related Articles | Metrics
Hyperspectral remote sensing is one of the frontier technologies in the field of remote sensing.It’s a hot topic in hyperspectral information processing to apply sparse coding model to process hyperspectral remote sensing image.To improve the accuracy of hyperspectral image classification,a hyperspectral remote sensing image classification method based on the second-moment spatial-spectral joint contextual sparse coding(SM-CSC) was proposed.First,a dictionary was obtained by training the samples selected from the ground-truth data,then the sparse coefficient of each pixel was calculated based on the learned dictionary.Afterward,the sparse coefficient was inputted to the classifier and the final classification result was obtained.The visible and near-infrared hyperspectral remote sensing image collected by Tiangong-1 in Chaoyang District of Beijing and the KSC hyperspectral image were applied to estimate the performance of the proposed approach.Comparisons with three other classification methods such as support vector machine(SVM),spectral sparse coding(SSC),and first-moment spatial-spectral joint contextual sparse coding(FM-CSC) were made.Experimental results show that the proposed method can yield the best classification performance with the overall accuracy of 95.74% and the Kappa coefficient of 0.9476 on the Tiangong-1 data and with the overall accuracy of 96.84% and the Kappa coefficient of 0.9646 on the KSC data.
Real-time Road Edge Extraction Algorithm Based on 3D-Lidar
LI Guang-jing, BAO Hong, XU Cheng
Computer Science. 2018, 45 (9): 294-298.  doi:10.11896/j.issn.1002-137X.2018.09.049
Abstract PDF(9552KB) ( 1748 )   
References | Related Articles | Metrics
A real-time road edge extraction algorithm based on 3D-lidar was put forward for the environmental perception of driverless cars.In this algorithm,the height feature points and the smooth feature points are extracted separately in the maps rasterized and layered from the lidar points cloud followed by the constraint of the road width to obtain the candidate edge points.Then the candidate points are polynomial fitted by the algorithm of random sample consensus(RANSAC).Finally,Kalman filter is used to predict and track the road edge.The experimental results show that the proposed algorithm can extract the edge of road in real time and robustly in both park and urban roads.What’s more,this algorithm has been applied successfully in 2017 World Intelligent Driving Challenge.
Research on Face Tagging Based on Active Learning
SUN Jin, CHEN Ruo-yu, LUO Heng-li
Computer Science. 2018, 45 (9): 299-302.  doi:10.11896/j.issn.1002-137X.2018.09.050
Abstract PDF(1455KB) ( 745 )   
References | Related Articles | Metrics
In the era of big data,tremendous images are available,whereas images with tags are sparse relatively.For the purpose of learning and research,it’s necessary to classify and annotate images,andmost images are relevant to faces,consequently face tagging is an effective tool to annotate images.However,the cost of manual annotation is high.Aiming at solving the problems of lacking tagged images and high manual annotation cost,a discriminative model based on the active learning inducing the posterior distribution over labels was proposed.The discriminative model is based on markov random field(MRF) and gaussian process(GP),and considers the objective connections between the positions and features of samples with the addition of match constraint and non-match constraint between samples.Match constraint means that samples have the same label,while non-match constraint means that samples have different labels.Experimental results indicate that choosing samples for manual annotation according to the posterior distribution over labels induced by the discriminative model can greatly improve the classification accuracy.
Energy-efficient Facial Expression Recognition Based on Improved Deep Residual Networks
DU Jin, CHEN Yun-hua, ZHANG Ling, MAI Ying-chao
Computer Science. 2018, 45 (9): 303-307.  doi:10.11896/j.issn.1002-137X.2018.09.051
Abstract PDF(3164KB) ( 767 )   
References | Related Articles | Metrics
To improve recognition rate and reduce power consumption of facial expression recognition systems,this paper proposed a facial expression recognition method using an improved deep residual networks(ResNets).Residual learning solves the degradation problem of the deep Convolutional Neural Networks(CNNs) to a certain degree and increases the network layers infinitely,but it makes deep CNNs face a more serious power consumption problem.To solve this problem,this paper introduced a new biologically-plausible activation function to improve ResNets and get a facial expression recognition method with both higher performance and lower power consumption.The Rectified Linear Units(ReLU) in the convolutional layers of ResNets are replaced with the new activation function Noisy Softplus.The obtained weights by using the improved ResNets can be directly applied to a deep Spiking Neural Networks(SNNs) architecture derived from the ResNets.The experimental results suggest that the proposed facial expression recognition method is able to achieve higher recognition rate and lower power consumption on a neuromorphic hardware.
Research on Low-resource Mongolian Speech Recognition Based on Multilingual Speech Data Selection
ZHANG Ai-ying
Computer Science. 2018, 45 (9): 308-313.  doi:10.11896/j.issn.1002-137X.2018.09.052
Abstract PDF(1702KB) ( 780 )   
References | Related Articles | Metrics
The performance of low-resource speech recognition system is improved by the multilingual information.However,when the multilingual information is used to improve the performance of low-resource automatic speech re-cognition system,notall of the multilingual speech data could be utilized to improve the performance of low-resource automatic speech recognition system.In this paper,a data selection method which is based on long short-term memory recurrent neural network based language identification was proposed and used to improve the performance of low-resource automatic speech recognition system.More efficient multilingual speech data are selected and used to train multilingual deep neural network and deep Bottleneck neural network.The deep neural network model obtained by using transfer learning and the Bottleneck features extracted by using the deep bottleneck neural network are both helpful to improve the performance of low-resource target language speech recognition system.Comparing with the baseline system,there are 10.5% and 11.4% absolute word error rate reductions under the condition of interpolated web based language mo-del for decoding.
Detection for Group Riot Activity Based on Change Analysis of Group Motion Pattern
HUANG Jin-guo, LIU Tao, ZHOU Xian-chun, YAN Xi-jun
Computer Science. 2018, 45 (9): 314-319.  doi:10.11896/j.issn.1002-137X.2018.09.053
Abstract PDF(2471KB) ( 1041 )   
References | Related Articles | Metrics
Group riot activity is the main precaution point of intelligent video surveillance because of its large hazards for social public safety.In view of the problem of low efficiency and low detection accuracy of the existing group riot activity detection algorithms,an activity detection algorithm for group riot activity based on change analysis of group motion pattern was proposed.This method extracts the optical flow features of foreground pixels as the basis of the activity analysis,and uses K-means clustering and Bayesian criterion to realize the grouping division of different groups in the scene.On this basis,it analyzes the changes in motion patterns of all groups,builds the maximum change factors,and computes the variation of the maximum change factors to detect the group riot activities.The experimental results show that this method used for detecting group riot activities has low false-alarm rate and miss-alarm rate and less average detection time.