15 April 2020, Volume 46 Issue 4
-
Select all|
-
Computer Engineering. 2020, 46(4): 1-10. https://doi.org/10.19678/j.issn.1000-3428.0056738Multi-Agent Path Planning(MAPP) is the problem of finding an optimal path set for multiple agents from their starting position to the target position without any collisions.Research on this problem has abundant application scenarios in the fields of logistics,military and security.This paper systematically sorts out and classifies the recent research progress in MAPP both at home and abroad.According to the differences in the optimality of results,the MAPP algorithms are classified into optimal group and approximate group.The optimal MAPP algorithms are mainly divided into four categories,which are based on A* search,cost growth tree,conflict based search and protocol respectively.The approximate MAPP algorithms are mainly divided into two categories:the unbounded suboptimal algorithm and the bounded suboptimal algorithm.Based on the classification mentioned above,this paper analyzes the characteristics of each algorithm,introduces the representative researches in recent years and discusses the future research directions of MAPP problem.
-
Computer Engineering. 2020, 46(4): 11-18. https://doi.org/10.19678/j.issn.1000-3428.0055169With the rapid growth and popularization of electronic devices and electric vehicles,how to guarantee the safety and stability of lithium-ion batteries becomes an important topic of relevant research,in which the Remaining Useful Life(RUL) of batteries becomes one of the most critical means to monitor the state of batteries.During the charge-discharge cycles,lithium-ion batteries undergo an irreversible process that can cause continuous degradation on battery capacity and end up in battery malfunction.In order to perform reasonable charge-discharge management that can meet the high reliability requirements in actual applications,this paper conducts a research on the RUL prediction in the using process of lithium-ion batteries.Four RUL prediction methods are expounded herein, which are based on mechanism model,data driven,mechanism and data driven fusion and data driven model fusion respectively,and the advantages and disadvantages of RUL prediction methods based on data driven are discussed.Moreover,the future research direction and trends are also summarized and predicted herein.
-
Computer Engineering. 2020, 46(4): 19-25. https://doi.org/10.19678/j.issn.1000-3428.0054989With the deep integration of Industrial Control System(ICS) and Internet technologies,it is important to detect system intrusion effectively for secure ICS.As network data of industrial control systems is high-dimensional and nonlinear,this paper applies Fisher score and kernel Principal Component Analysis(PCA) in preprocessing of network data.The standard Particle Swarm Optimization(PSO) algorithm tend to fall into local optimization in optimization of Support Vector Machine(SVM) parameters.To address the problem,a PSO algorithm based on Self-adaptive Mutation(SVPSO),is proposed to build a detection model for system intrusions.Simulation results on the standard dataset show that the detection model comstructed by SVPSO algorithm outperforms BPANN,KNN,random tree and naive Bayes algorithms in terms of detection performance,with the detection accuracy reaching 98.75% while the false alarm rate reduced to 1.22%.
-
Computer Engineering. 2020, 46(4): 26-32,39. https://doi.org/10.19678/j.issn.1000-3428.0055458With the rapid development of wireless networks,the problem of efficient allocation of spectrum resources in the Internet of Things(IoT) needs to be solved.Therefore,this paper proposes TSRA,a trust based spectrum resources allocation mechanism.By referring to the auction theory,the suction system model of spectrum resources is established.According to the trust theory,the trust relationship between users is determined to narrow down the scope of customer network and the transaction data is protected by the attribute encryption theory.On this basis,the improved ant colony algorithm is used to make reasonable plan for the resources allocation path,so as to achieve multi-objective allocation of spectrum resources.Experimental results show that the proposed mechanism can provide fine-grained protection for users' transaction data.It has high social benefits and low system computing and communication costs.
-
Computer Engineering. 2020, 46(4): 33-39. https://doi.org/10.19678/j.issn.1000-3428.0054879In the traffic field,the traditional traffic spatiotemporal data analysis method has drawbacks such low efficiency and poor reliability of spatiotemporal data similarity retrieval at congested intersections.From the perspectives of spatial scene similarity and dynamic data similarity,this paper proposes an intersection similarity calculation method based on spatiotemporal analysis.This method is comprised of a static data model and a dynamic model of intersections.The static data model takes the intersections as the spatial scene to calculate the spatial scene similarities between the target intersection and the database intersection.The dynamic data model builds a time series database according to the time attribute of the intersection detector data.Then the dynamic model slices the dynamic data of intersections in time dimension and calculates the similarities between the target intersection and the database intersection in the same period.On this basis,the similarity computing model is built and the intersections that satisfy the similarity retrieval are sorted to enhance the reliability of retrieval results.Experimental results show that compared with the spatial index retrieval algorithm,the proposed method can effectively improve the accuracy and efficiency of intersection retrieval.
-
Computer Engineering. 2020, 46(4): 40-45,52. https://doi.org/10.19678/j.issn.1000-3428.0054272In Chinese Named Entity Recognition(NER) based on neural network,the vectorized representation of words is an important step.Traditional representation methods for word vectors only map a word to a single vector,and cannot represent the polysemy of a word.To address the problem,this paper introduces the BERT pretrained language model to build a BERT-BiGRU-CRF model for representation of sentence characteristics.The BERT pretrained language model with bidirectional Transformer structure is used to enhance the semantic representation of words and generate semantic vectors dynamically based on their context.On this basis,the word vector sequence is input into the BIGR-CRF model to train the whole model,or train the BIGR-CRF part only with BERT fixed.Experimental results on MSRA data show that the F1 value in the two training modes of this proposed model reaches 95.43% and 94.18% respectively,which is better than that of the BIGRU-CRF,the RADICAL-BILSTM-CRF and the GRAIN-LSTM-CRF models.
-
Computer Engineering. 2020, 46(4): 46-52. https://doi.org/10.19678/j.issn.1000-3428.0054075The constraint satisfiability problem is a classic NP-hard problem,its basic algorithms are recursive backtracking and arc consistency algorithms.The combination of Arc Consistency(AC) and backtracking search can effectively reduce the size of solution space.Aiming at the maintaining problem,this paper proposes a new propagation scheme based on temporal counting to incrementally update the constraint subnet and use accumulateRevision and pushRevisionas the main process of two-way revision in order to reduce revision times and the numbers of domain filtering variables.Experimental results show that compared with the classic relationship-based scheme and variable-based propagation scheme,the overall solution speed of this scheme is significantly improved,and it has less time of revision.
-
Computer Engineering. 2020, 46(4): 53-59,69. https://doi.org/10.19678/j.issn.1000-3428.0054077To improve the efficiency of solving the Maximum Clique Problem(MCP) of large-scale legends on big data platforms,this paper proposes a maximum clique identification algorithm based on parallel constraint programming.The BMT graph division strategy is used to divide a complex graph into several sub-graphs,and the sub-graphs are assigned to computing nodes in the Spark cluster for independent calculation.Each computing node in the cluster uses the constraint programming method to model and solve the sub-problems generated by the segmentation.Thus,the parallel processing of the MCP can be realized.Secondly,a parallel graph partitioning method is designed based on the prediction model of task running time.So,the load balancing problem of computing nodes can be effectively solved.Experimental results show that compared with the identification method for maximum clique based on BMC graph division,the proposed algorithm is more efficient,and can obtain a nearly linear speed-up.
-
Computer Engineering. 2020, 46(4): 60-69. https://doi.org/10.19678/j.issn.1000-3428.0053661Existing Support Vector Guided Dictionary Learning(SVGDL) algorithm based on the principle of large-margin classification.When establishing decision-making hyperplanes,the algorithms consider only the boundary conditions of each class of encoding vectors,but ignore data distribution information,which limits the generalization ability of the model.To address the problem,this paper proposes a Minimum Class Variance Support Vector Guided Dictionary Learning(MCVGDL) algorithm.First,MCVGDL takes the Minimum Class Variance Support Vector Machine(MCVSVM),which combines Fisher linear discriminant analysis and the large margin classification principle of Support Vector Machine(SVM),as discriminant term.Second,during alternate optimization of model classifiers,MCVGDL comprehensively takes the distribution information of encoding vectors into account,to guarantee the overall consistency of encoding vectors of similar samples and reduce the coupling degree of corresponding components between vectors and modifies SVM classification vectors.So,the discriminant information of encoding vectors can be fully mined to better guide dictionary learning,improving the classification performance.Experimental results on face,object,and handwritten digit recognition datasets show that in terms of the recognition rate and atomic robustness,the proposed algorithm outperforms classical dictionary learning algorithms,including K Singular Value Decomposition(KSVD) and Local Constrained and Label Embedding Dictionary Learning(LCLE-DL),etc.
-
Computer Engineering. 2020, 46(4): 70-76,84. https://doi.org/10.19678/j.issn.1000-3428.0054276The texts in ecommerce usually do not follow the way of expression as the texts in general domains,resulting in low accuracy of traditional phrase mining methods in the ecommerce text mining.Therefore,this paper proposes a phrase mining method based on cooperative training.Through the phrase classification model based on semantic features,the antitone expression of ecommerce texts is effectively detected.Then the phrase mining framework of cooperative training is constructed,so as to reducing the cost of marking training data in the domain corpus.On this basis,the Stacking method is used to integrate the advantages of statistical model and semantic model,thus improving the overall mining performance of the model.Experimental results on Taobao query corpus show that compared with ClassPhrase and AutoPhrase methods,the proposed method has higher accuracy and recall rate.
-
Computer Engineering. 2020, 46(4): 77-84. https://doi.org/10.19678/j.issn.1000-3428.0054897Traditional text retrieval technology is mainly for one-dimensional text,and can hardly retrieve two-dimensional structural mathematical expressions.To solve the problem,a mathematical expression retrieval method based on operator information is implemented by introducing the Formula Description Structure(FDS).The method uses the extraction algorithm for FDS to obtain the node information of LaTeX mathematical expression,so as to obtain its skeleton storage structure.On this basis,the nodes with an operator value of 1 and their related documents are selected for indexing,and a set of expressions similar to the input is obtained by mathematical expression matching algorithm.Experimental results show that,this method can quickly and accurately find similar expressions from 519 588 mathematical expressions,and is not affected by the general operands.
-
Computer Engineering. 2020, 46(4): 85-90,96. https://doi.org/10.19678/j.issn.1000-3428.0054209In the recommendation system,the traditional Matrix Factorization(MF) cannot extract the features of users and items.The Neural Collaborative Filtering(NCF) adds a multi-layer perceptron to the model for this reason,but still cannot effectively make use of the auxiliary information other than the user and item ID.Therefore,this paper proposes a new conditional convolution method.In this method,the item features are used as input and the user features as convolutional kernels,so as to achieve the purpose of unshared weights,thus making the conditional convolution has stronger feature extraction and combination abilities while maintaining the features of parameters.On this basis,the conditional convolution can incorporate a variety of auxiliary information for personalized recommendation.Experimental results show that compared with the NCF model,the proposed method can increase the recommendation hit rate by 3.11% in the implicit feedback data and reduce the scoring prediction error by 2.47% in the explicit feedback data.
-
Computer Engineering. 2020, 46(4): 91-96. https://doi.org/10.19678/j.issn.1000-3428.0054056Naive Bayes(NB) algorithm is simple and efficient when applied to text classification,but it has a bottleneck in accuracy due to the intrinsic assumption that attribute independence and attribute importance are consistent.To solve this problem,this paper proposes a feature-weighted NB text classification algorithm based on Poisson distribution.The algorithm combines the Poisson distribution model with the NB algorithm,and the Poisson random variable is introduced into the weight of feature words.On this basis,the Information Gain Ratio(IGR) is defined to weigh the feature words of texts,and thus the effects of the attribute independence assumption of traditional algorithms can be reduced.Experimental results on the 20-newsgroups data set show that,compared with NB algorithm and its improved algorithms RW,C-MNB and CFSNB,this algorithm can improve the accuracy rate,recall rate and F1 value of text classification.Meanwhile,its execution efficiency is higher than K-Nearest Neighbor(KNN) algorithm and Support Vector Machine(SVM) algorithm.
-
Computer Engineering. 2020, 46(4): 97-106,122. https://doi.org/10.19678/j.issn.1000-3428.0053910Mining adjoint patterns of moving objects is finding the set of objects with similar trajectories and time from a spatio-temporal perspective,which is widely used in user behavior analysis based on geographical location.However,the existing similarity mining algorithms for moving objects can hardly address large amounts of temporally continuous and spatially discrete data of spatio-temporal correlation.To deal with the problem,this paper proposes a width-first search algorithm based on sliding window,Apriori property and greedy selection strategy to solve the problem of mining adjoint patterns of moving objects.Also,a two-layer pruning algorithm is designed combining the iterative pruning algorithm based on Hash iteration and pruning algorithm based on summary information,so as to remove redundant intermediate results.Experimental results on real data show that the pruning efficiency of the proposed algorithm is higher than that of pruning algorithms using only Hash iteration or summary information,meanwhile it can stably remove over 99% of redundant data.
-
Computer Engineering. 2020, 46(4): 107-114. https://doi.org/10.19678/j.issn.1000-3428.0053782When the user selects the most serious alarm in the existing interaction mode,it is completely based on his personal preferences,without considering the difference in the cost of processing different alarms.To this end,this paper proposes a weight search and alarm selection method based on user preference.In this method,the preference value of user's priority for alarm processing is explored,an evaluation function for the complexity of the problem is constructed and the selection strategy of preference weight is given.Then,the utility function is established for different alarms and their corresponding user preference weights,and the random threshold is set.Accordingly,the alarm selection scheme that needs to be solved first is determined.Under the constraint of cost,the selection is made based on the user's preference and the efficiency of alarm processing is optimized.Experimental results show that the method can reasonably and efficiently make alarm selection,and its performance of alarm selection is better than that of the backpack-based and threshold-based alarm selection method.
-
Computer Engineering. 2020, 46(4): 115-122. https://doi.org/10.19678/j.issn.1000-3428.0054701In order to meet the requirements of intrusion detectionfor real-time performance and accuracy,this paper designs an IL-SVM-KNN classifier that combines Support Vector Machine(SVM) and K-Nearest Neighbor(KNN) algorithm,and the balanced k-dimensional tree is used for data structure to improve the execution speed.In the training phase,the idea of incremental learning is applied and the expansion of the knowledge base is considered.In the classification phase,the SVM algorithm and KNN algorithm are used to divide the to-be-classified data into three cases,each case with a unique classification strategy.Experimental results on KDD CUP99 and NSL-KDD datasets show that the IL-SVM-KNN classifier can distinguish abnormal traffic from normal traffic,and determine the type of abnormal traffic attacks.The accuracy of the proposed classifier is significantly improved compared with the KNN algorithm and SVM algorithm.It also outperforms the decision tree,random forests and XGBoost algorithm in terms of the accuracy of determining the attack type while reducing the elapsed time and resource consumption compared with two-layer convolution neural network.
-
Computer Engineering. 2020, 46(4): 123-128,134. https://doi.org/10.19678/j.issn.1000-3428.0054476The network connection data can execute feature grouping according to the basic features of connection,the content features,the network traffic features and the host features.Taking advantage of this characteristic,this paper proposes a K-means based clustering method according to the grouping of features,so as to effectively achieve feature reduction and data dimensionality reduction.The differential information within the feature groups are retained by adjusting clustering parameters,and the decision tree C4.5 algorithm is used for intrusion classification of the data after dimensionality reduction.Experimental results show that the proposed method can reduce the number of clustering features of kddcup99 dataset from 41 to 4.The overall detection rate on network connection data is 99.73%,the false detection rate is 0 and the detection rates of normal network connection and Probe attack are both 100%.
-
Computer Engineering. 2020, 46(4): 129-134. https://doi.org/10.19678/j.issn.1000-3428.0054193Volunteer computing has been widely used for its openness,anonymity and dynamic features,but it also brings threats to system security.Traditional authentication methods cannot meet security requirements of volunteering computing systems,which can be solved by building the trust mechanism in systems.This paper proposes a trust model named VC-trust based on Bayesian theorem for volunteering computing systems.The uncertain behavior of nodes is analyzed and predicted based on Bayesian theorem.Then the trust values are calculated by introducing the punishment factor and adjustment function according to historical interactions reconds of nodes,and updated by using time sliding windows.Experimental results show that,in the case of node behavior changing,the VC-trust model has higher interaction success rate compared with BTMS model.
-
Computer Engineering. 2020, 46(4): 135-142,150. https://doi.org/10.19678/j.issn.1000-3428.0055801User behavior on a computer is directly reflected in the interactions with application windows.To address intranet security issues,research on user behavior is conducted from the perspective of the use of application windows.A completely free intranet environment is built,and user behavior data on application windows is collected and analyzed.On this basis,two kinds of behavior features of the use of application windows are extracted,which solve abnormal user detection and user change behavior recognition respectively.By using the sample mean distribution features and K-S test,it is verified that there are significant differences in the behavior of different users using application windows.Then,an abnormal behavior detection algorithm is constructed by combining Euclidean distance and confidence interval.Experimental results show that the algorithm can detect abnormal users and identify changed user behavior with a high accuracy.The accuracy rates are 97.4% and 94.5% respectively,which has practical application significance for preventing internal threats.
-
Computer Engineering. 2020, 46(4): 143-150. https://doi.org/10.19678/j.issn.1000-3428.0054698Many public audit schemes in public key cryptosystem have certificate management problem,which will increase storage load and communication cost.In order to effectively verify the integrity of the data in the semi-trusted cloud and reduce the extra cost of certificate management,this paper proposes a certificateless public audit scheme.The batch audit is realized by using homomorphic technology,so as to efficiently complete the audit needs of multiple users.The ELGamal encryption system is adopted to track user identity,thus preventing the malicious behavior of users.The results of security and performance analysis show that the proposed scheme is safe and efficient.It can resist typeⅠand typeⅡadversary attacks and satisfy the unforgeability of signature and the privacy of user identity.
-
Computer Engineering. 2020, 46(4): 151-156. https://doi.org/10.19678/j.issn.1000-3428.0054186For the management and control of network traffic,this paper proposes a parallel identification method for encrypted traffic based on the interaction fields of the Secure Socket Layer(SSL) protocol and Hidden Markov Model(HMM) with multiple inputs and a maximal single output.This method uses the fields at the interaction phase of the SSL protocol of the unidirectional data stream from the client or server as the observation sequence of a HMM,and forming a fingerprint database of HMM built for all to-be-identified encrypted applications.On this basis,the forward algorithm is used to calculate the probability of the unknown sequence being identified as HMM,and the application corresponding to the HMM with the highest probability is taken as the identification result.Experimental results show that the method has better identification performance and robustness for typical encrypted applications than traditional application identification methods.
-
Computer Engineering. 2020, 46(4): 157-161,182. https://doi.org/10.19678/j.issn.1000-3428.0054238The main security threats to Software Defined Network(SDN),a new type of network architecture,are from DDoS attacks.Hence,theestablishment of an efficient DDoS attack detection system is importantt to network security management.In the SDN environment,support protocols of existing DDoS intrusion detection algorithms are limited,and the algorithms have poor practicability.To address the problem,this paper proposes a DDoS attack detection algorithm based on Deep Belief Network(DBN).The DDoS attack mechanism in the SDN environment is analyzed.The SDN network topology through Mininet is simulated,and Wireshark is used for collection and detection of DDoS traffic data packets.Experimental results show that compared with XGBoost,random forest,and Support Vector Machine(SVM) algorithms,the proposed algorithm has excellent overall performance with high accuracy,low false alarm rate,fast detection rate and high easy scalability.
-
Computer Engineering. 2020, 46(4): 162-168. https://doi.org/10.19678/j.issn.1000-3428.0054313The existing malicious code classification method based on Convolutional Neural Network(CNN) has the problem of large computational resource consumption.In order to reduce the computational quantity and parameter quantity in the classification process,this paper constructs a malware family classification model based on malicious code visualization and lightweight CNN.The malware is visualized as grayscale to represent the similarity on code structure of the same malware family.Then the gray map is used to train the neural network model MobileNet v2 with deep separable convolution,so as to automatically extract the texture features.The Softmax classifier is used to classify the malicious code.Experimental results show that the average classificationaccuracy ofthe proposed model is 99.32%,which is 2.4 percentage points higher than the classic malicious code visualization model.
-
Computer Engineering. 2020, 46(4): 169-176. https://doi.org/10.19678/j.issn.1000-3428.0055179To solve the problem of the interference in the full bandwidth Beam Hopping(BH) satellite communication system,this paper proposes a BH pattern optimization method based on beam clustering.Slot allocation is made according to the proportion of each beam request to the available resources of the system,and the system capacity is improved on the premise of satisfying relative fairness.A distance threshold is calculated,and when the distance between beams is greater than the threshold,the influence of interference on signal quality can be ignored.Therefore,the purpose of avoiding interference is achieved by controlling the distance between simultaneous beams to be larger than the threshold.Simulation results show that compared with the traditional multibeam system,the resource allocation method increases the system capacity by 24.6%,eliminates the effect of system interference on signal quality,and improves the timeslot continuity of the BH pattern.
-
Computer Engineering. 2020, 46(4): 177-182. https://doi.org/10.19678/j.issn.1000-3428.0054253In the case that a tree-type anti-collision algorithm has too many tags,the efficiency of tag recognition is low due to the depth of the tree.Therefore,this paper proposes a tree anti-collision algorithm based on pseudo ID code.The reader uses the tag number prediction algorithm to detect the approximate number of unrecognized tags within the recognition range and send them to the tag.The tag randomly generates a number as its own pseudo ID code.The reader sequentially queries the pseudo ID code,and if a collision occurs,it uses the Collision Tracking Tree(CTT) algorithm for identification.The algorithm reduces the depth of the query tree and improves the recognition efficiency of the tag through the pseudo ID code in the process of identifying the tag.Theoretical analysis and simulation results show that compared with the CTT algorithm and the QT algorithm,the throughput rate of the algorithm is increased by 15% and 74% respectively.At the same time,the speed of tag recognition is effectively improved and the total number of timeslots is reduced.
-
Computer Engineering. 2020, 46(4): 183-188,197. https://doi.org/10.19678/j.issn.1000-3428.0055050In order to optimize the design and mapping of service function chains in Network Function Virtualization (NFV) and improve the utilization of physical resources,this paper proposes a service function chain design and mapping algorithm A-VNFC based on Virtual Network Function(VNF) combination.The design uses the Integer Linear Programming(ILP) model to find the optimal solution of the Total Bandwidth Consumption(TBC) of the objective function in a small-scale physical network.It searches for combinable VNFs,and uses the VNF decision tree to check all combination strategies,reducing TBC through iteration and optimization.Simulation results show that the proposed A-VNFC algorithm can effectively reduce bandwidth consumption in different scenarios,and its TBC value is close to the minimum bandwidth consumption value obtained by the ILP model.
-
Computer Engineering. 2020, 46(4): 189-197. https://doi.org/10.19678/j.issn.1000-3428.0054900To address imbalanced energy consumption in Wireless Sensor Network(WSN) and avoid death of overloaded "hot" nodes around the Sink node,this paper proposes an energy optimized routing algorithm for improved ant colony.In node distribution,the algorithm combines distance band and limited search angle with the distance factor to reduce energy consumption of nodes.The incentive mechanism is introduced to remove the "hot" nodes with insufficient residual energy and a longer path from the preferred path,while the nodes with fewer hops and sufficient energy are used to balance the transmission tasks of hot nodes.On this basis,a pseudo random proportional rule that includes energy factors is used to optimize the probability transfer function,which reduces the probability of hot node failure and enhances the optimization ability of the algorithm.Thus the algorithm can avoid falling into an untimely local optimum.Simulation results show that the proposed algorithm can effectively balance network energy consumption.Compared with the IEEABR and IARA algorithms,the algorithm has a longer network lifetime.
-
Computer Engineering. 2020, 46(4): 198-205. https://doi.org/10.19678/j.issn.1000-3428.0054300In the existing indoor passive location methods based on Channel State Information(CSI),the selection of sampling points has great impact on the feature matching accuracy and location accuracy of fingerprint database.According to the transmission features of WiFi and the fading features of channel,this paper proposes a 30° angle concentric circular sampling.In the offline phase,concentric circles are used to divide the detection area,taking samples every 30° angle.The differential signal features are extracted by the Principal Component Analysis(PCA) and the fingerprint database is constructed.In the online state,the land mobile distance algorithm is used for intrusion detection.When someone is detected,this method uses the improved Support Vector Regression(SVR) algorithm and introduces Gaussian kernel function to match the features of data,thus achieving precise location of the personnel.Experimental results show that compared with CSI-MIMI and FIFS methods,the proposed method has higher positioning accuracy and lower positioning error.
-
Computer Engineering. 2020, 46(4): 206-212,235. https://doi.org/10.19678/j.issn.1000-3428.0054561In order to solve the problems such as poor link stability,intermittent data transmission and so on in ultra wide band Wireless Sensor Networks(WSN),this paper proposes a link stabilization algorithm based on virus-antibody immune game.The area of nodes is evenly divided and further optimized by designing the coverage division method and combining the distance and residual energy factors,so as to reduce the jitter probability of links.The immune algorithm is introduced to construct a virus-antibody immune game mechanism based on the characteristics of the antibody between links and nodes,thus optimizing the clustering effect of nodes and links.The data interaction characteristics are improved through virus-antibody training,so as to improve the quality of links and the performance of regional transmission.Based on the energy-hop equilibrium method,the multi parameter decision mechanism is designed to evaluate the link connectivity performance of regional nodes and sink nodes,thus improving the control ability of the algorithm over link congestion.On this basis,the frequency domain orthogonal characteristic of PSK pre transmission method is used to set the transmission frequency for regional nodes one by one,so as to minimize the link jitter caused by frequency interference.Simulation results show that compared with LEACH and LMS-A algorithms,the proposed method has higher link stability,longer network stable running time and lower congestion frequency.
-
Computer Engineering. 2020, 46(4): 213-219. https://doi.org/10.19678/j.issn.1000-3428.0054066Due to the limited bandwidth,power and computational capability of sensor nodes in the Wireless Sensor Network(WSN),the traditional scheme hardly distinguish the abnormal data accurately when faced with massive data.To address this problem,this paper proposes an abnormal data detection scheme based on hierarchical distributed WSN.The scheme clusters the data at the node level by K-Means++ algorithm.Then cluster merging algorithm is used to reduce the amount of data transmission.The KNN-based abnormal cluster detection algorithm is performed on the gateway node,so as to return the normal cluster information to the underlying node for local detection,thus identifying the abnormal data.Experimental results on the Gaussian and IBRL datasets show that the detection rate of the proposed scheme is higher than 98%,and its communication consumption can be significantly reduced.
-
Computer Engineering. 2020, 46(4): 220-227. https://doi.org/10.19678/j.issn.1000-3428.0054582To address the inextensible fixed number of components in image prior modeling based on Gaussian Mixture Model(GMM),this paper proposes an extensible GMM model based on Dirichlet Process(DP).Through the addition and merging mechanism of cluster components,the model complexity can adaptively vary with data scaling,making the structure of the learnt prior model more compact to improve its interpretability.Also,to improve the inference of the proposed model,a scalable variational inference algorithm using batch processing is proposed to solve the variational distribution of all hidden variables for prior learning.Experimental results demonstrates that the proposed model has better denoising performance than EPLL and other traditional models with a higher Peak Signal-to-Noise Ratio(PSNR) in image denoising tasks,which proves its effectiveness.
-
Computer Engineering. 2020, 46(4): 228-235. https://doi.org/10.19678/j.issn.1000-3428.0054581Facial expression transfer is a key technology for character animation in computer vision,but existing facial expression transfer methods have some problems,such as unnatural expression generation,lack of realism,complex transfer model and difficulty in training.Therefore,a face expression transfer model based on conditional Generative Adversarial Network(GAN) is constructed.The condition of the expression domain is specified by the classification loss function of the design domain,so that a single generator can learn the mapping relations between multiple expression domains.Meanwhile,the conditional constraints and zero-sum game between the model generator and the discriminator are used to realize the transfer of five facial expressions by training only one generator.Experimental results show that,this model can effectively transfer facial expressions and has strong robustness.Facial expressions generated by the proposed model are more natural and realistic than the StarGAN model.
-
Computer Engineering. 2020, 46(4): 236-240,246. https://doi.org/10.19678/j.issn.1000-3428.0054665In order to ensure the stereo matching efficiency of SAD algorithm and improve the matching accuracy,this paper proposes an Edge-Gray algorithm with edge features.The edge feature map is obtained by edge calculation.During the matching process,the size of the matching window and the source map are determined according to the difference between the current point and the field point.On this basis,the disparity map is obtained by performing parallax optimization.Experimental results show that compared with the traditional SAD algorithm,the average mismatch rate of the Edge-Gray algorithm is lower,it has best stereo matching effect for Cones images with more edges,with a mismatch rate reduced by 10.52%.
-
Computer Engineering. 2020, 46(4): 241-246. https://doi.org/10.19678/j.issn.1000-3428.0054319Measurement matrix is an important part of the compressed sensing theory,and is related to the reconstructed accuracy of original signals.To address the low reconstruction accuracy of common measurement matrixes,this paper proposes a random measurement matrix based on Markov Chain.Firstly,M random numbers are generated by using the randomness of Markov Chain,and respectively mapped to -1 and 1 based on certain rules to serve as the elements of an M×M diagonal matrix.Secondly,M×(N-M) random numbers are generated by using Markov Chain,and mapped to 0 and 1 respectively to form an M×(N-M) matrix.Finally,the two matrixes are combined to form an M×N measurement matrix.Simulation results show that the proposed matrix has a simple structure,and can significantly improve the reconstruction accuracy compared with commonly used measurement matrixes and Toeplitz structure matrixes based on singular value decomposition.The proposed matrix also reduces the amount of calculation and saves storage space.
-
Computer Engineering. 2020, 46(4): 247-252,259. https://doi.org/10.19678/j.issn.1000-3428.0054245To address the low speed and accuracy of existing semantic segmentation networks,this paper proposes a fast scenario semantic segmentation method based on the dense layer and attention mechanism.The method adds the dense layer and attention module into ResNet.The dense layer adopts two-channel transmission that helps in obtaining multi-scale targets.It also uses group convolution to reduce the amount of computation.An attention module is introduced into the feature extraction network to reduce the loss of accuracy.Experimental results show that the proposed method gets an MIOU of 81.5% and a speed of 42.3 frame/s on the cityscape dataset.It also gets an MIOU of 61.8% and a speed of 27.9 frame/s on the ADE20K dataset,which means the proposed method can improve the segmentation speed while keeping the accuracy.
-
Computer Engineering. 2020, 46(4): 253-259. https://doi.org/10.19678/j.issn.1000-3428.0056382In order to better extract the global features of smoke images,this paper proposes a smoke recognition method based on dilated convolution and dense connection.The method stacks in order the expansion convolutions with different expansion rates to expand the receptive field of the convolution kernel,so the convolution kernel can perceive a wider area of smoke images.The dense connection mechanism is designed between different dilated convolutional layers to promote the information exchanges between layers,and realize the fusion of local and global features of smoke images.On this basis,a deep convolutional neural network is constructed for smoke recognition,and is trained on the convex combination of training samples and labels to enhance the generalization ability of the model.Experimental results show that compared with methods such as AlexNet and VGG16,this method has better smoke feature expression performance,and can achieve more reliable smoke recognition effect with a smaller model,which proves its excellent practicability.
-
Computer Engineering. 2020, 46(4): 260-265. https://doi.org/10.19678/j.issn.1000-3428.0054626Due to the differences of imaging principle between visible images and infrared images,pedestrian detection algorithms for visible images cannot be directly applied to infrared images.To address the problem,this paper proposes a pedestrian detection algorithm for infrared images based on multi-level features.First,the key regions of infrared images are extracted by using an improved image saliency detection algorithm,and their highlighted regions are rapidly located by using the sliding window algorithm for centroid relocation.Then the symmetry of images and their similarity to pedestrian features are judged by using Zernike moment.Finally,the regions to be judged are gradually narrowed down by using the Convolutional Neural Network(CNN) based on edge information input.Experimental results on the OTCBVS dataset of infrared pedestrian images show that compared with the sparse representation algorithm,the proposed algorithm has a higher detection accuracy rate.
-
Computer Engineering. 2020, 46(4): 266-272. https://doi.org/10.19678/j.issn.1000-3428.0056246Due to the difficulty of obtaining depth cues in some special scenarios,the application of existing 3D content generation methods is limited.Therefore,by replacing depth map with saliency map for 2D to 3D conversion,this paper proposes a 3D content generation method.The rough saliency map is generated by using Fully Convolutional Network(FCN) and the output results are optimized by Conditional Random Field (CRF).Experimental results show that the proposed method can solve the problem of low quality of saliency map in existing methods caused by using low-level features for visual attention analysis,and it can generate 3D content with satisfying visual effect.
-
Computer Engineering. 2020, 46(4): 273-278,286. https://doi.org/10.19678/j.issn.1000-3428.0053998The method of determining adaptive parameters is not clearly given in existing second-order multi-agent systems,and the convergence speed of systems is low.To predict the state of a multi-agent system for aircraft at the next moment in practical applications and improve the convergence speed,this paper proposes a parameter adaptive consensus algorithm.The algorithm takes the difference of location and speed between current agents as the feedback parameter of the consistency protocol.On this basis,the finite-time consistency problem of a second-order multi-agent system in the case of fixed topology and switching topology is studied.The Lyapunov function is constructed,and the LaSalle's invariant set principle and homogeneity theory are used to obtain the conditions required for the system to reach stability in finite time,so as to implement adaptive adjustment of input states of different aircrafts.Simulation results show that the proposed algorithm can ensure consistent tracking of multi-agent systems in a finite-time,and the convergence speed is fast.
-
Computer Engineering. 2020, 46(4): 279-286. https://doi.org/10.19678/j.issn.1000-3428.0054360In view of the lack of real-time and efficient method and monitoring technology for watershed heavy metal ecological risk assessment,this paper uses remote sensing and meteorological data to construct a reasonable and accurate evaluation model to monitor the environmental conditions of the decision basin.By collecting meteorological remote sensing and soil heavy metal data in the study area and combining the Hakanson potential ecological risk index,this paper proposes an Ecological Risk Assessment (ERA) model for watershed heavy metals.Using Microsoft Visual Studio 2013,ArcGIS and other development platforms and WebGIS technology to jointly complete the B/S architecture system development,this paper achieves remote assessment management of the watershed ecological environment.The application results show that the system realizes the functions of heavy metal content analysis,ecological risk classification and evaluation in the study area and increases the real-time performance of the ecological risk assessment of the study area and the accuracy of the data accumulation result analysis.
-
Computer Engineering. 2020, 46(4): 287-293,300. https://doi.org/10.19678/j.issn.1000-3428.0054605The current detection accuracy rate of crowd abnormal behaviors in public places is low,and most abnormal behaviors such as sudden running cannot be recognized.Therefore,this paper proposes a recognition algorithm for crowd abnormal behaviors based on YOLO_v3 and sparse optical flow,providing sufficient time for early warning and corresponding emergency measures for crowd anomalies by detecting small group anomalies.In order to locate the abnormal area conveniently,the video is divided into several sub-areas and the image samples of sub-areas are obtained to detect small group anomalies that induce crowd abnormality.The improved YOLO_v3 neural network is adopted to detect anomalies that are difficult for the traditional algorithms to detect,such as carrying stick,gun and knife and face occlusion.When these anomalies are not detected,the sparse optical flow method is used to obtain the average kinetic energy and the entropy of the movement direction of the crowd,and the obtained characteristic data is sent to PSO-ELM for classification,distinguishing normal behaviors from co-directional spur or irregular spur.Experimental results show that compared with existing similar algorithms,the proposed algorithm can effectively detect small group anomalies such as pedestrian armed anomalies and facial occlusion anomalies,and can locate the area where the abnormality occurs with an accuracy rate of 98.227%.
-
Computer Engineering. 2020, 46(4): 294-300. https://doi.org/10.19678/j.issn.1000-3428.0054599The safety of electric transmission line is the prerequisite of a safe and stable operation of power grid,but the damage caused by birds directly threatens the safe operation of electric transmission line.To address the disadvantages of traditional bird repellent start-stop strategy,this paper proposes a bird detection model for electric transmission lines based on YOLO v3 algorithm.This model obtains the image data through electric transmission line monitoring device,extracts the deep features of images by residual module and uses multiple scale object detection strategy to guarantee the bird detection effect.Experimental results show that in the bird detection tasks for electric transmission line,the accuracy of the proposed model can reach 86.75% and the detection speed can be up to 47 frame/s.This model can accurately and timely detect the bird number around the electric transmission line and its high robustness is verified in rainy,foggy and jittering scenes,which proves it can guarantee a safe and stable operation of electric transmission line.
-
Computer Engineering. 2020, 46(4): 301-308,315. https://doi.org/10.19678/j.issn.1000-3428.0053149Multiplex network is a graph structure in which multiple link relations exist between nodes.When constructing similarity index,most of existing multiplex network link prediction algorithms consider only the topological attributes of a single-dimensional network,and fails to mine the relations between sub-networks of different dimensions,which undermines the performance of link prediction.To address the problem,this paper proposes a multiplex network link prediction algorithm based on multiplex global node influence identification index,Multiplex PageRank(MPR).By defining a multiplex node influence ranking index MPR,the nodes with greater influence in the multiplex network space can be measured.Then,the influence ranking function is converted into the score of similarity between two nodes in each potential node pair,and applied to the multiplex network link prediction scene.Experimental results on two real multiplex network datasets show that the proposed algorithm outperforms PR,EDC,ANC and other algorithms,and has better stability.
-
Computer Engineering. 2020, 46(4): 309-315. https://doi.org/10.19678/j.issn.1000-3428.0054160Currently,intelligent information processing of the Zhuang language is in its fancy and lacks automatic tagging methods for parts of speech.To address the lack of Zhuang corpus,arduousness of manual tagging,and poor performance of machine tagging,this paper proposes a part of speech tagging method for the Zhuang language based on reinforcement learning.The method builds a tag dictionary according to the grammatical features of Zhuang and Chinese Penzhou Tree Bank(CTB) symbols,and uses dependency syntax analysis to fuse semantic features.Then Long Short-Term Memory(LSTM) network serves as strategic network,using cyclic memory to improve part of observation information.On this basis,a reinforcement learning framework is introduced,and the target part of speech is used as environmental feedback.The true value of the target is gradually approached through feature learning.Experimental results show that this method has excellent performance in part of speech tagging of Zhuang.It can alleviate the dependency of the part of speech tagging model on training corpus,and enlarge the tag dictionary of Zhuang language quickly.Besides,the proposed method can realize part of speech tagging of Zhuang language automatically.
-
Computer Engineering. 2020, 46(4): 316-320. https://doi.org/10.19678/j.issn.1000-3428.0054556In the deep learning based speech enhancement model,the Long Short-Term Memory Network(LSTM) can well handle the sequence speech enhancement problem,but the training speed of the model is slow when dealing with speech enhancement problems based on large-scale noisy speech data.Aiming at this problem,this paper proposes a speech enhancement method based on quasi Recurrent Neural Network(RNN).The gate functions and memory cells are used to ensure the correlation of the context of noisy speech sequences,and the calculation of gate functions is no longer dependent on the output of the previous moment.Moreover,the model introduces the convolution operation of the matrix in the calculation of the input of the noisy speech sequence and gate functions,so that the model can simultaneously process the speech sequence information at multiple moments,thereby enhancing the parallel computational ability.Experimental results show that compared with the LSTM,the proposed method can greatly improve the training speed of the network model under the premise of ensuring the speech enhancement performance.
2024
Volume 50 No 3
Volume 50 No 3
(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