15 August 2019, Volume 45 Issue 8
-
Select all|
-
Computer Engineering. 2019, 45(8): 1-6. https://doi.org/10.19678/j.issn.1000-3428.0054221Threat adjudication based on the judge method of ruling difference is an important mechanism for the mimic defense system to shield and block the threat of attacks.However,the existing mimic adjudication mechanism cannot conduct effective inductive analysis and threat control on the security situation of the mimic defense systems.Therefore,taking the mimic Web service system as an example,and integrating the network situation awareness technology into the mimic defense architecture,this paper proposes an improved Web threat situation analysis method.The data association is performed on the multi-level mimic adjudication alarm log.The feature data information extracted by fusion is deeply mined and classified.Different types of classification data are visually displayed.Experimental results show that the method can display the security state of the mimic defense systems,and is informed of the running state of the abnormal execution body in time,so as to realize the analysis and evaluation of the security situation of the mimic defense systems.
-
Computer Engineering. 2019, 45(8): 7-13,21. https://doi.org/10.19678/j.issn.1000-3428.0051747This paper proposes a algebraic fault attack method based on optimized fault location against SIMECK cipher.By analyzing encryption diffusion defect of the SIMECK round function and the failure cause,the deterministic propagation characteristics of faults are extracted,and the differential characteristic table of deterministic faults is constructed to achieve accurate fault location.It creates an equivalent equation set for the encryption process and fault information,and converts the equations into SAT problems and solve the key.Experimental results show that the method can inject a random single-bit fault into the left register in the 28th round of SIMECK32/64,and only need 8 fault injections to recover the complete 64 bit master key.The attack success rate is 99.61%.Compared with the existing fault attack method,the proposed method requires fewer fault samples and has a higher attack success rate,creation of equations more automated.
-
Computer Engineering. 2019, 45(8): 14-21. https://doi.org/10.19678/j.issn.1000-3428.0053425Scheduling is an important mechanism for the Web server with mimic structure.Most of the existing scheduling algorithms lack consideration about heterogeneity and Quality of Service(QoS) of the Web server with mimic structure,and do not solve the problems of security and service quality instability caused by the scheduling mechanism.Therefore,a scheduling algorithm called Random Seed algorithm based on Maximum heterogeneity and Web QoS(RSMHQ)is proposed.All the thresholds of the Web servers with mimic structure are calculated.The seed executor is randomly selected,and the scheduling scheme is determined according to the maximum heterogeneity and QoS.Simulation results show that compared with the random scheduling algorithm,the proposed algorithm has better scheduling effects and achieves an excellent balance between security,Web service quality and dynamic behaviour.
-
Computer Engineering. 2019, 45(8): 22-24,30. https://doi.org/10.19678/j.issn.1000-3428.0051008In order to construct samples that conform to language norms in fuzzy testing of language interpreter,and get abnormal test results as far as possible to find vulnerabilities,the improved Probabilistic Context Free Grammar(PCFG) model is used to control the variation process of samples,and the undefined variables in the variation results are modified to increase the ratio of samples that conform to language norms.On this basis,the language interpreter is tested by fuzzy testing.Results show that the ratio of samples generated by the test that conform to the grammatical and semantic norms is as high as 96%.
-
Computer Engineering. 2019, 45(8): 25-30. https://doi.org/10.19678/j.issn.1000-3428.0055156Aiming at the problem that an attacker can issue pirated software by re-signing the vulnerabilities in the self-signature process of Android system application program,this paper analyses the digital signature and verification process of Android system application program,proposes an online signature comparison scheme based on MD5 value,and increases the difficulty of attacker's decompilation through code obfuscation technology to ensure the security of the signature scheme.Experimental results show that the scheme can help users to judge whether the application is authentic or not,and can effectively prevent the local signature from being tampered with.
-
Computer Engineering. 2019, 45(8): 31-34,41. https://doi.org/10.19678/j.issn.1000-3428.0051218In order to achieve accurate identification of malware encrypted C&C communication traffic,this paper analyzes the https communication process of normal Webpage browsing access and C&C communication,discovers the server independence feature of malware C&C communication and proposes a sequence modeling method of https communication.Based on the behaviour characteristics of encrypted communication,a vector representation method for hexadecimal characters of ciphertext is used to implement a vectorized expression of encrypted traffic.Multi-window Convolutional Neural Network(CNN) is used to extract the pattern characteristics of encrypted C&C communication and realize the identification and classification of encrypted C&C communication data traffic.Experimental results show that the accuracy of identifying the encrypted C&C communication traffic of malware is 91.07%.
-
Computer Engineering. 2019, 45(8): 35-41. https://doi.org/10.19678/j.issn.1000-3428.0052117In order to study the impact of the difference of input rates and removal rates of nodes in the network on worm virus propagation,this paper analyzes the transformation relationship between each compartment based on the idea of compartment modeling and constructs a worm virus propagation SEIR model with different input rates and removal rates.It calculates the equilibrium point and basic regeneration number of the model,gives the stable specification of equilibrium point,and uses the Hurwitz theorem,the LaSlle invariance principle and the Bendixson theorem to prove the stability of equilibrium point.On this basis,this paper verifies the results of theoretical analysis by numerical simulation and analyzes the key factors that affect the worm virus propagation,then proposes some suggestions and measures to suppress the worm virus propagation.
-
Computer Engineering. 2019, 45(8): 42-47. https://doi.org/10.19678/j.issn.1000-3428.0051519To meet the migrating needs of applications based on the Docker container between Kubernetes clusters and Docker Swarm clusters,an application migration system for A-Migrator heterogeneous container cloud is developed.Through the application orchestration and scheduling strategies of Kubernetes and Docker Swarm,the paper gives a feasible method of converting orchestration information between the two and the technology of application migration based on pre-synchronization of images.Experimental results show that the application migration method for A-Migrator heterogeneous container cloud can migrate applications based on Docker container between Kubernetes clusters and Docker Swarm clusters successfully,and the time used to migrate applications decreases by 60.33% on average after introducing the technology of image pre-synchronization.
-
Computer Engineering. 2019, 45(8): 48-52,59. https://doi.org/10.19678/j.issn.1000-3428.0051542In order to perform qualitative combination clustering analysis on mixed data sets containing numeric and noun attributes,a qualitative combination algorithm f-QRD for rough clustering based on fuzzy dominance relations is proposed.The mixed data sets are clustered according to different attributes and calculating the fuzzy dominance relations between clusters.In order to avoid the excessive fragments of clusters after combination,clusters with small difference of fuzzy dominance relations are merged.Experimental results show that compared with QRD algorithm,f-QRD algorithm can effectively reduce the number of clusters and improve the clustering accuracy.
-
Computer Engineering. 2019, 45(8): 53-59. https://doi.org/10.19678/j.issn.1000-3428.0052493In order to display three-dimensional massive data in real time,this paper proposes an improved rapid display algorithm for massive data.The CURE clustering algorithm is used to sort the data,and the data is indexed by Hilbert R-tree.The visual area prediction model predicts the next time visible area to realize rapid visualization of large amounts of data.Experimental results show that compared with the Visualization algorithm based on Motion of Viewpoint (VMV) and the Visualization algorithm based on Testing of Visibility (VTV),the rendering speed is 18.27% higher than the VMV algorithm without reducing the rendering quality.Compared with the VTV algorithm,the increase is 67.06%,the prediction area error rate is reduced by 9.73% compared with the VMV algorithm,and the VTV algorithm is reduced by 22.37%,which can quickly load data and accurately draw three-dimensional large data volume.
-
Computer Engineering. 2019, 45(8): 60-65,74. https://doi.org/10.19678/j.issn.1000-3428.0053141Mobile edge computing can improve service quality and reduce network load,but existing edge resources cannot meet all of the user demands.To increase the resource utilization and service benefits,an edge network service auction method based on robust optimization.This paper studies how users participate in the resource allocation process by means of auction,considering the uncertainty and credibility security of users' bids.The method also takes edge providers' profits from obtained resources based on auction results,so as to find the resource allocation strategy that leads to maximum benefits.Analysis results show that the method can effectively improve the profits of limited edge resources and reduce the hammer price.
-
Computer Engineering. 2019, 45(8): 66-74. https://doi.org/10.19678/j.issn.1000-3428.0052286In the selection of Single Nucleotide Polymorphism(SNP) from high-dimensional genetic data with few samples,in order to make the selected SNP subset highly represent all SNP information and achieve data dimension reduction,an improved method is proposed on the basis of Fuzzy C-Mean(FCM) algorithm,which is named GN-FCM.By introducing the weight factor of SNP,the difference of importance degree of SNP site is quantified.Meanwhile,the neighborhood regular term of key SNP is introduced into the loss function of fuzzy clustering,so as to mine the correlation between highly important SNP and other SNPs in the neighborhood.Experimental results show that GN-FCM has better convergence.Compared with DW-FCM algorithm,the accuracy of the constructed SNP subsets by this algorithm in Support Vector Machine(SVM),Decision Tree(DT) and Naïve Bayesian(NB) classification is improved by 5.73%,3.40% and 3.79% respectively,and the F1 value is improved by 4.01%,3.20% and 2.22% respectively.
-
Computer Engineering. 2019, 45(8): 75-79,91. https://doi.org/10.19678/j.issn.1000-3428.0051759In order to optimize the classification effect of unbalanced data,an improved fuzzy decision tree algorithm is proposed combining the hesitant fuzzy set theory and the decision tree algorithm.The unbalanced data is oversampled by the SMOTE algorithm,the cluster center point of each attribute is obtained by using the K-means clustering method,and the datasets is fuzzy processed by using two different membership functions.On this basis,the Hesitant Fuzzy Information Gain(HFIG) of each attribute is obtained by the information energy of hesitant fuzzy sets and membership functions.The largest HFIG is used to replace the FIG in the Fuzzy ID3 algorithm as the split criterion of the attribute,and a Hesitant Fuzzy Decision Tree(HFDT) model is constructed for unbalanced data classification.Experimental results show that,compared with traditional classification algorithms such as C4.5,KNN and random forest,the classifier based on HFDT has an average increase of 12.6% on the AUC evaluation index.
-
Computer Engineering. 2019, 45(8): 80-85. https://doi.org/10.19678/j.issn.1000-3428.0053297To address the datasets of Eclipse project software defect reports of Bugzilla defect tracking system,feature selection and machine learning algorithms are used to perform feature dimension reduction and weight optimization on vectorized original data to obtain optimized dataset with lower data dimension,and classification algorithms are used to evaluate the severity of software defect reports.The cross-comparison results of the four feature selection algorithms and the four machine learning algorithm results show that the Information Gain(IG) feature selection algorithm is used to perform feature optimization on the original dataset,and the optimized dataset is trained and tested by using the Multinamial Naive Bayes(MNB) algorithm.The AUROC value of the severity prediction of software defect reports can be increased to 0.767.
-
Computer Engineering. 2019, 45(8): 86-91. https://doi.org/10.19678/j.issn.1000-3428.0052616Aiming at the application requirement of microwave radiometer for small and medium scale meteorological prediction,a data measurement and control system for meteorological microwave radiometer is proposed.The system takes the STM32F4 series ARM processor as the core and Altera Cyclone IV E series FPGA as the auxiliary.By designing the overall structure of the system,the peripheral circuits of ARM processor and FPGA chip,and the corresponding embedded software,the functions of data collection,communication between upper and lower computer and voltage detection are realized.In order to improve the precision and time resolution of the follow-up inversion,the combination of FPGA and ARM processor is used to synchronously collect the 16-channel voltage output of the receiver.Experimental results show that the system can accurately control the antenna,the state of noise source and the constant temperature switch,and has good real-time performance.
-
Computer Engineering. 2019, 45(8): 92-101. https://doi.org/10.19678/j.issn.1000-3428.0051265In order to improve the penetration testing efficiency of an Industrial Control System(ICS),ensure its security and reliability and enhance the system security protection ability,this paper constructs a penetration testing tool framework for ICS based on shell interaction technology and implements it with the Python language.It designs a hierarchical network detection module and system detection module,and uses protocol parsing and reverse technology for vulnerability testing of industrial control protocols.It also researches vulnerability exploitation ways based on the industrial control environment and designs a fuzzy testing module for vulnerability mining and detection of testing targets.On this basis,a penetration attack script is written according to the template rules in reference to the open source software,Metasploit.Simulation results show that this design can improve the detection efficiency for ICS,reduce the difficulty of protocol vulnerability detection.Meanwhile,it has the characteristics of simple structure and easy extension.
-
Computer Engineering. 2019, 45(8): 102-106,112. https://doi.org/10.19678/j.issn.1000-3428.0052669The scanning,processing and identification processes of existing iris login systems are under the same physical machine,which leads to the problems of high use cost and few benefited users.Therefore,an iris login system model is proposed.This model separates the scanning front-end and the processing back-end through the network,and places the iris scanning front-end in the browser and the processing back-end on the cloud server.The front-end technology is used to build user registration and identify login window.At the same time,multiple users can use the browser to call the optimized iris cameras to obtain their personal iris information and send it to the cloud server.On the cloud server end,Django and Nginx are used to deploy the iris algorithm application,and MySql is used to store user information.The CPU core is used to speed up image processing and parallel search through multi-threading and multi-process technologies.Experimental results show that this model improves the processing speed of one hundred images by 60% and the image matching search efficiency by 75% on average compared with the ordinary iris recognition process.
-
Computer Engineering. 2019, 45(8): 107-112. https://doi.org/10.19678/j.issn.1000-3428.0053128According to the performance and functional requirements of the principle prototype of optical synthetic aperture,a computation and control system composed of Field Programmable Gate Array(FPGA) and Digital Signal Processor(DSP) is designed.The FPGA is responsible for multi-channel high-speed data collection and parallel pipeline preprocessing,and the DSP completes parallel computation of data and realizes closed-loop control.After the closed-loop of the computation and control system,real-time collection and preprocessing of three channels images are carried out simultaneously,and the relevant computation and real-time closed-loop control are completed.Results of open-loop and closed-loop experiments on the prototype show that the root mean square of image offset of the three sub-telescopes in the closed-loop system is only 5% of that in the open-loop system,which can ensure the stable interference imaging of the sub-telescopes.
-
Computer Engineering. 2019, 45(8): 113-119. https://doi.org/10.19678/j.issn.1000-3428.0051847Due to the tight coupling of the traditional Airborne Tactical Network(ATN) in the fight application without considering the information content itself,the routing protocol for the ATN cannot provide differentiated communication services for different information content,and it is hard to meet the aviation cluster platform's communication requirements,which has dynamic and various task requirements.Therefore,this paper combines the Software Definition Network(SDN),Information Center Network(ICN) and airborne tactical network architecture,puts forward the Software Defined Information Center-Airborne Tactical Network(SDIC-ATN) architecture.Based on this architecture,the Content-driven Routing Protocol(CDRP) is designed,which can provide different levels of differentiated routing forwarding abilities for different information content,and provides efficient differentiated services for network.Simulation results show that compared with the traditional Dijkstra algorithm,CDRP can reduce the end-to-end delay to some extent,improve the throughput and improve the network communication performance.
-
Computer Engineering. 2019, 45(8): 120-124,134. https://doi.org/10.19678/j.issn.1000-3428.0053485Considering the task granularity and Decoupling Level(DL),an improved software crowdsourcing task decomposition algorithm is proposed.The task granularity is calculated based on the dependency relationship in the task network.The decoupling level is measured according to the distribution of each subtask in the Design Structure Matrix(DSM),and the software crowdsourcing task decomposition is performed by dynamic decoupling.Experimental results show that compared with the task decomposition algorithm based on Independent Level(IL) and Propagation Cost(PC),the risk judgment value and defect density of the algorithm are increased by 0.244 0,0.362 6,0.014 6 and 0.319 4 respectively,which can ensure the completion quality of software crowdsourcing tasks.
-
Computer Engineering. 2019, 45(8): 125-128. https://doi.org/10.19678/j.issn.1000-3428.0051639The local repair code is applied in distributed storage system.In this code,errors in any bit of code word can be repaired by reading other bits of the code word.According to this characteristic,the research is carried out around the three-dimensional and four-dimensional optimal codes.By discussing the relevant parameters of the known special optimal codes,analyzing the linear relationship between the column vectors of the generator matrix of the known optimal codes,and using matrix transformation,matrix concatenation,matrix subtraction and other methods,all the three-dimensional and four-dimensional optimal codes in the 5-ary domain are constructed.On this basis,the local repair degree of the code is analyzed as small as possible,and the optimal local repair degree of the code is determined by the C-M boundary,so as to obtain the local repair degree with optimal distance.
-
Computer Engineering. 2019, 45(8): 129-134. https://doi.org/10.19678/j.issn.1000-3428.0051120Based on the Delay Tolerance Remaining Time(DTRT),an uplink priority scheduling algorithm for dynamic grouping of traffic is proposed.A Resource Block(RB) allocation method with variable size is adopted to satisfy the adjacency constraints of RB,and a hybrid automatic polling mechanism is introduced to solve the unfair scheduling problem of low priority traffic.Simulation results show that the system throughput of this algorithm is increased by 15% and 41% compared with PF algorithm and RR algorithm respectively,and it has better performance in terms of traffic delay and traffic differentiation.
-
Computer Engineering. 2019, 45(8): 135-140. https://doi.org/10.19678/j.issn.1000-3428.0051676The Wireless Local Area Network(WLAN) adopting the IEEE 802.11 protocol cannot provide differentiated quality of service to different types of data streams,and the IEEE 802.11e service differentiation mechanism EDCA exists the unfairness problem in data stream transmission.Therefore,this paper uses the NS-3 network simulator to implement an improved service differentiation mechanism F-DCF.A queuing rule is given for each individual node to ensure fair transmission opportunities within the node,and a fair channel access mechanism between multiple nodes is designed and implemented by using a contention based access method.Simulation results show that compared with the traditional EDCA mechanism,the F-DCF mechanism can provide proportional fairness of service quality differentiation for different priority data flows in the network,and can improve the total throughput of the network.
-
Computer Engineering. 2019, 45(8): 141-145. https://doi.org/10.19678/j.issn.1000-3428.0051935Aiming at the massive Multiple Input Multiple Output(MIMO) systems based on pilot pollution,a channel estimation algorithm with Space-Alternating Generalized Expectation-Maximization(SAGE) iteration is proposed.A complete data space set is formed with the transmitted and received pilot symbols.The pilot-based Minimum Mean Square Error(MMSE) estimator is used to initialize the channel parameters.Pilot symbols and soft information iteration is used to improve and update initial estimated data.Experimental results show that the algorithm can achieve precision convergence through a SAGE iteration,and it has higher estimation accuracy and spectral efficiency than the existing pilot-based estimator in the case of high Signal Noise Ratio(SNR) and pilot pollution.
-
Computer Engineering. 2019, 45(8): 146-151. https://doi.org/10.19678/j.issn.1000-3428.0051591The cooperative combat of mixed formation of Unmanned Aerial Vehicle(UAV) and manned aircraft can improve combat performance,but requires a stable data connection between the UAV and manned aircraft to transmit battlefield situation to the manned aircraft quickly.To address the problem,a single price and multi-attribute bidding model based on game theory is proposed.The selection process of the transmission node is abstracted into the bidding model by using the bidding method of neighbor nodes in the communication range.Based on the node energy attribute,link stability and forwarding angle,the scoring function and the revenue model of the bidding node are designed,so that the bidding node finally selects the best node to complete the data forwarding by comparing the quality attributes of the bidding nodes and the bidding price.Game analysis proves that the algorithm is incentive-compatible,ensuring that nodes can participate in the data forwarding process with real quotes.Simulation results show that compared with GPSR algorithm and AMIM algorithm,the algorithm can guarantee network energy consumption balance and link stability while improving network lifetime.It is suitable for cooperative combat environment of mixed formation of UAV and manned aircraft.
-
Computer Engineering. 2019, 45(8): 152-158,164. https://doi.org/10.19678/j.issn.1000-3428.0051753Nodes in Ad Hoc networks are prone to show selfish behavior when forwarding data.To motivate selfish nodes to participate in data forwarding,a nodes incentive strategy IMTFT is proposed.Based on Bayesian game theory,a nodes forwarding game model is established,in which an improved TFT strategy with additional incentive factor is introduced to equally motivate selfish nodes.Under the IMTFT strategy,the Nash equilibrium conditions of nodes are reasoned and analyzed,and the optimal parameters of the incentive factor are determined.Simulation results show that the proposed strategy can effectively motivate selfish nodes to participate in data forwarding and improve the overall performance of the networks.
-
Computer Engineering. 2019, 45(8): 159-164. https://doi.org/10.19678/j.issn.1000-3428.0051934Power Line Communication(PLC) is an important communication mode for the internet of things and smart grids.However,the new nodes cannot recognize the specific modulation mode used in PLC networks when try to access the networks,which causes the failure of accessing.Therefore,by constructing the PLC Multiple Input and Multiple Output(MIMO) system model,a cooperative modulation recognition method based on PLC-MIMO structure is proposed for new nodes to access PLC networks automatically.The fourth-order cumulants using the signal transmitted from multiple nodes are taken as the feature parameters,and the decision rule of the maximum likelihood is involved to execute cooperative recognition.At the same time,an improved cooperative modulation recognition algorithm is designed to improve the correct recognition rate.Simulation results show that compared with the single node recognition algorithm and the general collaborative recognition algorithm,the correct recognition rate of the algorithm is improved by 16% and 6% respectively,which has better recognition performance and is suitable for PLC sensor network system.
-
Computer Engineering. 2019, 45(8): 165-172,189. https://doi.org/10.19678/j.issn.1000-3428.0051757Aiming at the characteristics of large number of users,high concurrent communication and time-varying in the space-ground integrated network,a model-driven distributed satellite terminal user behavior simulation method is proposed based on multi-scale virtualization technology.The multi-scale virtualization technology is used to effectively overcome the contradiction between simulation fidelity and computational overhead,and large-scale simulation of satellite terminals are realized.Using the model-driven design method,the theoretical simulation model can be combined with network simulation to achieve high-fidelity simulation of user behavior characteristics.Experimental results show that the proposed method can accurately realize large-scale simulation of 1 million online users and 100 thousands concurrent calls,and can accurately reflect the time-varying features of user behavior.
-
Computer Engineering. 2019, 45(8): 173-177,197. https://doi.org/10.19678/j.issn.1000-3428.0051208To optimize resource allocation in imperfect spectrum detection environment,a distributed cognitive radio network resource allocation algorithm is proposed.Based on Bayesian theory,the concepts of subcarrier state trust index and statistical average interference power are given.By using Lagrange dual decomposition theory,the original allocation problem is decomposed into independent sub-problems and solved to achieve effective resource allocation for multiple users.Simulation results show that compared with the optimization algorithm based on avoidance criterion,the proposed algorithm can effectively improve the system capacity and achieve the end condition of the cycle faster.
-
Computer Engineering. 2019, 45(8): 178-183. https://doi.org/10.19678/j.issn.1000-3428.0051516The Chinese real-word error in the online learning community will make it difficult to understand the semantics of Chinese texts,which affects the learning and analyzing effects based on online learning community texts.To this end,this paper proposes a real-word error detection and repairing method for short texts in online learning communities.Firstly,the confusion word set and the fixed collocation knowledge base corresponding to the confusion word are automatically constructed.Then,n-gram scores,context scores and fixed match scores are calculated for each confusion word based on the n-gram probability statistical model,context model,and fixed collocation knowledge base respectively.Finally,the weighted summation is used as the basis for judging whether the original text is wrong,and the confusing word with the highest score is used as the repair opinion.Experimental results show that this method can effectively detect and repair Chinese real-word error in the learning community,whose Recall,Precision,and Correction are 85.6%,86.3%,92.9% respectively.
-
Computer Engineering. 2019, 45(8): 184-189. https://doi.org/10.19678/j.issn.1000-3428.0053034Traditional collaborative filtering algorithms only rely on low-dimensional vector method of user rating data,which results in low accuracy of recommendation results and cold-start problems.Therefore,this paper proposes a new dynamic hybrid recommendation algorithm,which integrates the stacked denoising Autoencoder(AE) into the user-based collaborative filtering to learn users'deep features,and is integrated with the similarity based on user item attribute preference.In the prediction generation stage,the time attenuation item is set to dynamically predict the accessing probability and update the changes of user interest in time,so as to improve the quality of recommendation.Experimental results on the MovieLens dataset show that the recommendation performance of the algorithm is significantly improved compared with the UB-CF,AE,and SDAE-IA algorithms.
-
Computer Engineering. 2019, 45(8): 190-197. https://doi.org/10.19678/j.issn.1000-3428.0051610Most algorithms for online Multi-Task Learning(MTL) constrain task relatedness via a single weight matrix,and the constraint is so strict that it can not always hold in practice.Thus an improved online MTL algorithm is proposed to overcome this constraint by decomposing the weight matrix into two sub-matrics.A trace-norm regularization is imposed on the first sub-matrics to induce a low-rank correlative structure.A group Lasso penalty over individual tasks is applied on the second sub-matrics through a regularization term to identify personalized patterns.The projection gradient algorithm is applied to the adaptive learning of these sub-matrices and the optimal solution is obtained.Experimental results show that the proposed algorithm can achieve a sub-linear regret with respect to the best linear model in hindsight,the accuracy and the running time of the algorithm is better than TRML、MTFL and other algorithms,its cumulative error rate on the spam mail data set can be reduced to 4.97%.
-
Computer Engineering. 2019, 45(8): 198-202,209. https://doi.org/10.19678/j.issn.1000-3428.0051294The traditional movie recommendation algorithms mostly make recommendations based on the static attributes of users and movies,which ignores the inherent temporal and causal relationship in the time series data.The recommendation quality need to be improved.To this end,taking the advantage of Recurrent Neural Network (RNN) in processing time series,this paper proposes a recommendation algorithm R-RNN.Two Long-Short Term Memorys (LSTMs) are used to mine the potential status of users and films,long historical status accumulation is realized,and the prediction score is obtained by the inner product of the user status and the movie status.Experiments on IMDB and Netflix datasets and Netflix subsets show that compared with those of the probability matrix decomposition,TimeSVD++ and AutoRec algorithms,the algorithm can effectively reduce the root mean square error and improve the accuracy of the rating prediction.
-
Computer Engineering. 2019, 45(8): 203-209. https://doi.org/10.19678/j.issn.1000-3428.0051987In Location-Based Social Networks(LBSNs),real-time recommendation data of Point-of-Interests(POIs) and check-in data of users are highly sparse.Therefore,a hybrid recommendation model based on time effect is proposed.Through the data model of potential POIs of users,the time behavior influence scores and geographical location influence scores are calculated,and the linear unified model is used for processing.Top S POIs are selected as the potential POIs of users.Incorporate the user's potential check-in recorders into a time-based matrix factorization model,taking into account the influence of time difference and continuity on the recommendation results.On this basis,optimization solution is carried out,and the recommended strategy is proposed.Experimental results show that compared with the LRT model and UTE+SE model,the proposed model has better recommendation effect,and its precision and recall rate can reach up to 0.103 4 and 0.111 8.
-
Computer Engineering. 2019, 45(8): 210-216,223. https://doi.org/10.19678/j.issn.1000-3428.0051810In order to enhance the text semantic capture ability of text matching model and improve the semantic matching accuracy,a text matching model based on word embedding and dependency relation is proposed.It constructs the semantic representation of the fusion of word semantic and dependency relation between words,and obtains the matrix describing the semantic matching degree of each part of the two texts by cosine mean convolution and K-Max pooling operation.The Long-Short Term Memory(LSTM) network is used to learn the mapping relationship between the matching degree matrix and the true matching degree.Experimental results show that the F1 value of the model is 0.927 4,which is more accurate than the BM25 and deep text matching models.
-
Computer Engineering. 2019, 45(8): 217-223. https://doi.org/10.19678/j.issn.1000-3428.0052473This paper analyzes the part of speech and syntactic relationship between the evaluation object and the evaluation phrase in the commodity review,and proposes a method for evaluating collocation extraction using rule template.The core collocation extraction rules are designed by the results of part of speech,dependency syntactic analysis and semantic dependence analysis.The COO algorithm and the improved ATT chain algorithm are introduced,and the rule template for identifying the complete evaluation object and phrase is further developed according to the part of speech of the core evaluation object and the phrase to extract the evaluation information.Experimental results on the Chinese commodity review dataset show that compared with the nearest distance method,SBV polarity transfer method and core sentence-based method,the F1 value of the proposed method is increased by 43.98%,36.30% and 24.83% respectively.
-
Computer Engineering. 2019, 45(8): 224-229. https://doi.org/10.19678/j.issn.1000-3428.0051535In order to predict the personalized aesthetic preferences of specific objects,machine learning and depth learning are combined,and a personalized facial attractiveness evaluation method that integrating consensus and subjective preference information is proposed.Group ratings are collected from the data set to form consensus information,and the prediction model of individual ratings is constructed to reflect subjective information.Combining these two kinds of information and giving full play to the universality of consensus information and the uniqueness of subjective information,a personalized facial attractiveness prediction model is constructed.Experiments on open SCUT-FBP and FaceScrub datasets show that the Pearson correlation coefficient of this method is as high as 0.90 and the residual value is only 0.25.
-
Computer Engineering. 2019, 45(8): 230-235. https://doi.org/10.19678/j.issn.1000-3428.0053924Aiming at the situation that the area to be resigistered in the Augmented Reality(AR) system is occluded when the virtual information is superimposed,a three-dimensional registration method for AR system based on real-time feature detection and target tracking is proposed.KCF algorithm is used to perform real-time FAST feature matching on the target area and the template.By improving the ORB algorithm,SIFT-ORB-MRANSAC fusion algorithm is proposed to extract and match feature remove mismatches.On this basis,the registration matrix is calculated by the matching relationship between the feature points,and virtual information is overlaid to the registration matrix to augment the real world.Experimental results show that the proposed method can quickly and accurately overlay the virtual information to the registered area under the conditions of illumination change,occlusion and rotation transformation,and improve the points as well as computation efficiency of AR systems.
-
Computer Engineering. 2019, 45(8): 236-240. https://doi.org/10.19678/j.issn.1000-3428.0052795In order to reduce the computational complexity of Screen Content Coding(SCC),a fast algorithm for Coding Unit(CU) partitioning and Prediction Unit(PU) mode selection process is proposed.According to the depth information of the adjacent CU in the space-time domain and the average rate distortion cost,early determination is made for CU partition.The PU mode selection process is terminated based on the rate distortion cost means of the merge mode.Experimental results show that compared with SCM6.0,the algorithm saves 35.4% of the coding time when the average bit rate is only increased by 1.59%,which effectively reduces the computational complexity of the coding.
-
Computer Engineering. 2019, 45(8): 241-247. https://doi.org/10.19678/j.issn.1000-3428.0051612To address the poor tracking performance of the LCT algorithm when the target is in deformation or fast moving,a tracking algorithm based on feature fusion is proposed.Based on the Histogram of Oriented Gradient(HOG) feature correlation filtering,the target and background color histogram features are extracted to obtain the target prediction position of the color feature.On the basis,the feature fusion weight is determined according to the tracking confidence.Tracking results are obtained by considering gradient features and color features.Experimental results show that compared with the LCT algorithm,the distance accuracy and overlap accuracy of the algorithm are increased by 11.5% and 21.2%,respectively,and the average center position error is reduced by 15.3 pixels.
-
Computer Engineering. 2019, 45(8): 248-254. https://doi.org/10.19678/j.issn.1000-3428.0053232Extracting the Mel-Frequency Cepstral Coefficients(MFCC) in speech emotion recognition will lose the spectral feature information,resulting in a low accuracy of emotion recognition.Therefore,a speech emotion recognition method combining MFCC and spectrogram features is proposed.The MFCC features are extracted from the audio signal,the signal is converted into a spectral map,and the image features are extracted using a Convolutional Neural Network(CNN).On this basis,the Multiple Kernel Learning(MKL) algorithm is used to fuse the audio features,and the generated kernel functions are used to support the vector machine for emotion classification.Experimental results in two kinds of speech emotion data sets show that the speech emotion recognition accuracy of this method is as high as 96% compared with the classifier based on single feature.
-
Computer Engineering. 2019, 45(8): 255-259. https://doi.org/10.19678/j.issn.1000-3428.0051780The traditional speech enhancement method based on Deep Neural Network (DNN) has a fixed delay in processing due to its non-causal input,which is unsuitable for the real-time applications.To solve this problem,studying from the perspective of network structures,comparing the speech enhancement performance of different network structures under different input formats through experiments,the network structure suitable for the causal input is found in this paper.On this basis,by combining Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM),a causal speech enhancement model that can fully utilize the information of previous frames is established.Experimental results show that the proposed model is able to improve the real-time performance of the DNN-based speech enhancement method while ensuring the speech enhancement performance,whose PESQ and STOI scores are 2.25 and 0.76.
-
Computer Engineering. 2019, 45(8): 260-265,274. https://doi.org/10.19678/j.issn.1000-3428.0052107To meet the robustness,uniqueness and compactness requirements of video copy detection systems,a video fingerprint algorithm including temporal and spatial information is proposed.The temporal and spatial information is composed of spatio-temporal slices and keyframes.Weighted fusion is implemented on Gabor features including the spatial information of video key frames,and on histogram features including temporal information of spatio-temporal slices.The fusion is quantized to generate the video fingerprint.Comparative experiments are carried out on the public database,and results show that the proposed algorithm has an outstanding ROC performance,obviously higher robustness,and better overall performance compared with structure diagram model,temporal information image,gradient direction centroid and other algorithms.
-
Computer Engineering. 2019, 45(8): 266-274. https://doi.org/10.19678/j.issn.1000-3428.0052593The tracking algorithm based on Kernelized Correlation Filter(KCF) has strong dependence on the spatial structure of the target,cannot effectively deal with the occlusion,deformation and other interference factors,and the single feature model is unable to accurately represent the target information in complex tracking occasions.Therefore,an adaptive multi-model tracking algorithm is proposed.The correlation filter model and the color histogram model are combined by adaptive weight,and the idea of sparse representation is introduced into the training process of the correlation filter model to further enhance the robustness of the algorithm.Experimental results on the OTB dataset show that the algorithm can effectively mitigate the interferences from various factors such as occlusion and deformation in the tracking process,and compred with Staple algorithm,KCF algorithm and others,the algorithm achieves significant improvement in target tracking accuracy.
-
Computer Engineering. 2019, 45(8): 275-280. https://doi.org/10.19678/j.issn.1000-3428.0053695Insulators are an important part of transmission lines,and their normal operation directly affects the stable operation of the power grid.Therefore,intelligent insulator defect detection method is researched.The dataset is made by the aerial image of the drone,the K-means++ algorithm is used to determine the a priori frame,and an improved lightweight network is built based on the YOLOV3 detection architecture.Experimental results show that the method improves the image detection speed of high-definition insulators and can complete insulator positioning and defect detection.
-
Computer Engineering. 2019, 45(8): 281-286,295. https://doi.org/10.19678/j.issn.1000-3428.0051790The features extracted by the static analysis method based on feature code matching often lag behind the development of the virus and cannot detect unknown viruses.Therefore,the feature extraction and fusion are carried out from the virus decompilation file and theirs grayscale image.The Random Forest(RF) algorithm in machine learning is used to classify the malware code family,extract the two local features of the opcode instructions and grayscale texture of the malware code,and use the color histogram as the global feature of the malware code.Experimental results show that the fusion of malware code features and RF algorithm can achieve effective classification of malware code families,and the average accuracy rate is 99.59%.
-
Computer Engineering. 2019, 45(8): 287-295. https://doi.org/10.19678/j.issn.1000-3428.0051603Using the static behavior characteristics of social network users to identify spammer users,it is impossible to detect the dynamic behavior of spammer users and it is difficult to apply to online detection.Therefore,by constructing the dynamic behavior characteristics of social network users,the differences between normal users and spammer users are analyzed.Based on the semi-supervised model,the online detection model is constructed by using dynamic and static behavior characteristics.The data is filtered by using static behavior characteristic clustering and dynamic behavior filtering.The semi-supervised model uses the most valuable unlabeled user data for incremental learning to detect spammer users.Experimental results show that the F1 value of the model is as high as 93.33% and the average training time is about 2 minutes,which can effectively detect the spammer users on social network.
-
Computer Engineering. 2019, 45(8): 296-299,308. https://doi.org/10.19678/j.issn.1000-3428.0051468In order to give consideration to the safety and efficiency of scalar multiplication in resource limited security chip,an Elliptic Curve Cryptography(ECC) scalar multiplication algorithm resisting power analysis attacks is presented.The Non-Adjacent Form(NAF) coding method with threshold is used to encode the scalar to improve the efficiency of scalar multiplication.On this basis,the technologies of precomputation and base point masking are used to make the algorithm can resist the power analysis attacks.Analysis results show that the proposed algorithm can not only resist the Simple Power Analysis(SPA) attack,Differential Power Analysis(DPA) attack,Zero-value Power Analysis(ZPA) attack and Refined Power Analysis(RPA) attack,but also can improve the efficiency of precomputation and reduce storage space.
-
Computer Engineering. 2019, 45(8): 300-308. https://doi.org/10.19678/j.issn.1000-3428.0051617Based on the classification and analysis of various types of data involved in the two-dimensional and three-dimensional modeling of overground and underground oilfields,a hybrid model is built. The model consists of overground CSG model, surface GRID model, and underground Pillar model, providing unified data structure and interfaces for the overground and underground data. Visual modeling methods are analyzed, such as three-dimensional terrain, three-dimensional oil reservoir, three-dimensional surface features, 2D-to-3D transformation of surface vectors, and conversion of coordinate systems and coordinates. Based on the OSG platform, a unified three-dimensional scene is constructed. A plane-based three-dimensional mesh blanking algorithm and a neighborhood force mesh smoothing algorithm based on discrete attribute values are proposed. The three-dimensional vertical sectioning, unfolding of sections, and projections of oil well trajectories are studied. The horizontal section design optimization techniques of horizontal wells based on the genetic algorithm is implemented. This paper takes an oil well in Inner Mongolia as the study object, and results show that this system can be used for three-dimensional integrated modeling of digital overground and underground oil fields.
-
Computer Engineering. 2019, 45(8): 309-314. https://doi.org/10.19678/j.issn.1000-3428.0052437The Android platform malicious software detection method based on machine information extracts the feature information with high feature dimension and weak class distinguishing ability,resulting in low detection accuracy and high complexity.Therefore,a malicious software detection method based on Feature Ratio Difference(FRD) and weighted Random Forest(RF) is proposed.By obtaining the permission information and hardware component information of the Android software,the ration difference of various features is analyzed,and the feature attribute is used as the input of the classification model.On this basis,different weights are assigned to the tree model in the Random Forest(RF),and the impact of the tree model on the final classification result is verified.Experimental results show that the features extracted by the feature construction method based on the feature difference ratio have better class distinguishing ability,and the improved random forest can improve the accuracy of malware detection.
-
Computer Engineering. 2019, 45(8): 315-320. https://doi.org/10.19678/j.issn.1000-3428.0051831Online review text has the characteristics of colloquialism,and its evaluation words lack corresponding evaluation objects,which affects the effect of fine-grained sentiment analysis.Therefore,a method of automatic identification of evaluation objects based on deep learning is proposed.The specification of text sequence annotation in the research field is designed,the Named Entity Recognition (NER) of the evaluation word,as well as the evaluation object after the comment word segmentation is annotated,and the word sequence,the word part sequence and the tagging sequence are achieved.Transfer the word sequence and the word part sequence to the word vectors of the neural network language model,and train them with Recurrent Neural Network (RNN).Conditional Random Field(CRF) is used to output the tag of the evaluation object,and the evaluation object is achieved as well.Experimental results show that compared with single CRF model,BiLSTM+CRF model and BiGRU+CRF model have better recognition effects and the F1 value of BiGRU+CRF model can reach up to 0.84.
2024
Volume 50 No 4
Volume 50 No 4
(Started in 1975, Monthly)
Sponsored by：
East-China Institute of Computer Technology Shanghai Computer Society
Edited & Published by：
Editorial Board of Computer Engineering
Chief Editor：Jiang Bo
Executive Chief Editor:
Hu Bing
Tel：021-67092217
E-mail: ecice06@ecict.com.cn
CN: 31-1289/TP
ISSN: 1000-3428
Sponsored by：
East-China Institute of Computer Technology Shanghai Computer Society
Edited & Published by：
Editorial Board of Computer Engineering
Chief Editor：Jiang Bo
Executive Chief Editor:
Hu Bing
Tel：021-67092217
E-mail: ecice06@ecict.com.cn
CN: 31-1289/TP
ISSN: 1000-3428