15 September 2019, Volume 45 Issue 9
-
Select all|
-
Computer Engineering. 2019, 45(9): 1-7. https://doi.org/10.19678/j.issn.1000-3428.0052883Existing research on location prediction usually neglects the correlation between time and space.To address the problem,this paper proposes a Long and Short Term Memory model based on Spatial and Temporal features(ST-LSTM).Based on the LSTM network,a spatial-temporal gate that independently processes the spatial-temporal information of the user's move is added,and the spatial and temporal factors of the user's sign-in are added,so that the model has spatial-temporal characteristics.The personal correction factor is introduced in the ST-LSTM network to correct the output of each type of users,highlighting the individuality on the basis of ensuring the basic characteristics,and better learning the behavioral trajectory characteristics of each type of users.At the same time,a simplified variant model of two ST-LSTM networks is proposed under the premise of ensuring the characteristics of the ST-LSTM network.Results of testing on the public dataset show that compared with the mainstream location prediction methods,the prediction model has better performance in terms of accuracy,recall rate,and F1 value.
-
Computer Engineering. 2019, 45(9): 8-16. https://doi.org/10.19678/j.issn.1000-3428.0054115To explore the underlying interconnection structures among entities or objects in relational databases through emerging graph systems,this paper designs and implements a data pipe tool called R2GPipe to convert relational data to graph data.A concise declarative Domain-Specific Language(DSL) is given to describe the mapping relationship between relational elements and graph elements.Depending on analysis needs,users write mappings from relations to property graphs by R2G mapping language.R2GPipe generates code for the source system and the target system by parsing the R2G mapping language.Through the case study of the dataset TPC-H,the relational data is modeled as graph data to test the scalability of R2GPipe.Results show that the overall running time of R2GPipe increases linearly with the increase of the scale of the converted data.
-
Computer Engineering. 2019, 45(9): 17-22. https://doi.org/10.19678/j.issn.1000-3428.0052483To address the problem of data replica loss in cloud data center environment,two kinds of Quality of Service(QoS) aware data replica recovery algorithms are proposed.Based on the distribution of data replicas,it divides the data security level on the premise of considering the data replica security rules,and a phased data replica recovery algorithm is given.On this basis,according to how often the data is accessed or transited,the data replica rarity model is established to realize the rarity-aware phased data replica recovery algorithm.Experimental results show that compared with the direct data replica recovery algorithm,the proposed algorithms can effectively improve the application service quality and data security guarantee ability of the cloud data center.
-
Computer Engineering. 2019, 45(9): 23-31,39. https://doi.org/10.19678/j.issn.1000-3428.0053513Sparse Matrix-Vector Multiplication(SpMV) calculation in sparse matrix storage format is inefficient,and the computing results of Blocked Row-Column(BRC) storage format lack reproducibility and certainty.To solve the problem,this paper proposes an improved Blocked Row-Column Plus(BRCP) storage format.The BRCP storage format adopts different two-dimensional blocking strategies,adaptively adjusts the blocking parameters according to the statistical characteristics of the distribution of non-zero elements of each row in the matrix,and improves the parallelism of SpMV on the GPU platform.A GPU kernel fuction based on fast segmented summation algorithm is designed to ensure the certainty of calculation results and their reproducibility on different GPU platforms.Experimental results show that the BRCP storage format has high computational efficiency,which reduces the SpMV calculation error in the parallel environment and improves the accuracy of PageRank sorting compared to the BRC storage format.
-
Computer Engineering. 2019, 45(9): 32-39. https://doi.org/10.19678/j.issn.1000-3428.0053331In big data applications,most modeling methods are based on a complete data set,but data missing in the data acquisition process or storing process tend to result in failure to modeling.Therefore,a clustering-based recursive filling method is proposed.The incomplete data is pre-filled using the mean of the same cluster to form an initial complete data set.The complete data obtained are clustered,and the initial filling is corrected using the mean of the same cluster.The filling stability is determined according to the deviation of filling results,and the filling value is corrected through multiple times of recursive clustering until the last two times of filling is stable or the number of iterations exceeds the threshold.Experimental results show that compared with the methods of mean filling,K nearest neighbor filling,cluster filling and incomplete data analysis for rough sets,the method can implement more precise filling,making the final filling more close to real data.
-
Computer Engineering. 2019, 45(9): 40-48. https://doi.org/10.19678/j.issn.1000-3428.0052687The scientific workflow deployment in the cloud environment is different from the traditional independent task scheduling,and the scheduling time and cost should be considered simultaneously.To address the problem,a scientific workflow scheduling method based on budget allocation is proposed.The mapping between workflow tasks and virtual machine resources is divided into two stages:budget allocation,and resource provision and scheduling.In order to optimize budget usage,a budget allocation algorithm based on fast-priority,called FFTD,and budget allocation algorithm based on slow-priority,called SFTD,are designed to achieve sub-allocation of budget among tasks.The task selection is performed based on the descending order of the earliest completion time of the task,and the resources are allocated according to the sub-budget of the single task when the virtual machine is reusable,thereby ensuring smooth scheduling of the workflow task.Five kinds of conventional types of scientific workflows are introduced to test the performance of the algorithm under different types of workflow structures and different budget constraints.The results show that the FFTD algorithm has shorter scheduling time and higher virtual machine resource utilization and satisfaction rate of budget constraints than the BDT-AI algorithm in 72%,88%and 84% experimental scenarios,and the overall performance is better.
-
Computer Engineering. 2019, 45(9): 49-54. https://doi.org/10.19678/j.issn.1000-3428.0052133In order to solve the problem that the system Bit Error Rate(BER) is too high due to data interference between paired users in the Non-Orthogonal Multiple Access(NOMA)downlink,a NOMA scheme based on data correlation between paired users is proposed.The transmitting terminal estimates the phase rotation angle of the user sequence according to the non-correlation of the paired user data sequence,and designs a signal processing method for adaptive selection of the threshold.It sends the phase rotation mark and the superimposed signal of the user bit sequence to the user through the Gaussian channel,and the decoding method is selected at the receiving terminal by using a rotation flag carried by the downlink control information.Simulation results show that the scheme can reduce the mutual interference between the superimposed signal constellation points on the constellation diagram,and effectively improve the system BER performance and user fairness under the premise of adding a small amount of calculation.
-
Computer Engineering. 2019, 45(9): 55-59. https://doi.org/10.19678/j.issn.1000-3428.0052052In the Time Difference of Arrival(TDOA) positioning technology,when the position coordinates of the sensor have errors,the estimated value obtained by the Least Squares(LS) method no longer has the best unbiasedness,resulting in a decrease in the direction finding accuracy.Aiming at this problem,a direction finding method based on Total Least Squares(TLS) is proposed.By transforming the nonlinear observation equation into pseudo-linear equation,the augmented matrix is constructed and the singular value decomposition is carried out to achieve the target position.Theoretical analysis and simulation results show that this method is more accurate than the classical LS method and LS-Tylar method.
-
Computer Engineering. 2019, 45(9): 60-64. https://doi.org/10.19678/j.issn.1000-3428.0051896In order to realize all-IP communication between Wireless Sensor Network(WSN) and IPv6 Internet,an all-IP WSN address configuration scheme is proposed.The WSN is divided into multiple clusters to design a distributed IPv6 address configuration scheme.The cluster head node uses the stateful address allocation method and joins the dynamic tree of access node to obtain the IPv6 address.The cluster member node obtains an IPv6 address from the cluster head node within one hop through stateless address allocation,and performs duplicate address detection only within the scope of the cluster.Simulation results show that compared with the 6LoWPAN WSN hierarchical address configuration scheme,the proposed scheme has lower address configuration cost and delay.
-
Computer Engineering. 2019, 45(9): 65-69. https://doi.org/10.19678/j.issn.1000-3428.0053794Aiming at the problem of limited energy and unbalanced consumption of nodes in Wireless Sensor Networks(WSN),a utility function considering the residual energy of nodes,the success rate of information transmission and the distance between nodes and base stations is designed to construct a game model by using potential game theory.On this basis,a non-uniform topology control algorithm BLTC based on potential game is proposed.Simulation results show that,compared with DIA algorithm and VGEB algorithm,BLTC algorithm can balance the energy consumption of nodes and prolong the network life cycle.
-
Computer Engineering. 2019, 45(9): 70-75. https://doi.org/10.19678/j.issn.1000-3428.0051983In order to improve the efficiency of node scheduling in Heterogeneous Directional Sensor Network(HDSN),a Differential Evolution(DE) algorithm with parameter self-adaption based on learning automata is proposed.It transforms the node scheduling problem into the set covering problem,and realizes adaptive selection of differential algorithm's control parameters by using interaction between learning automata and environment.Meanwhile,it adopts the adaptive mutation strategy to enhance the optimization ability of the algorithm in solving the set coverage problem.Simulation results show that,compared with original DE algorithm,the proposed algorithm expands the parameter's self-adaptability and has stronger optimization ability.Meanwhile,it can prolong the lifespan of HDSN.
-
Computer Engineering. 2019, 45(9): 76-81. https://doi.org/10.19678/j.issn.1000-3428.0052038In order to construct high-dimensional core matrix polar codes quickly,a two-stage polar code construction method TPMC based on Monte Carlo(MC) is proposed.In the first stage,the most reliable and unreliable bits are obtained by using the Gauss approximation method with linear complexity.In the second stage,the above bits are fixed as frozen bits and MC method is implemented to measure the error rate of the remaining bit channels.The bits with a lower error rate are selected from the remaining bits,and are combined with most reliable information bits in the first stage to form an information bits set.Simulation results show that compared with MC method,TPMC method can reduce the computational complexity and improve the decoding efficiency.
-
Computer Engineering. 2019, 45(9): 82-86,94. https://doi.org/10.19678/j.issn.1000-3428.0052102In the wireless Energy Harvesting (EH) cooperative communication system,the energy bottleneck may occur at the relay.For this problem,an energy-saving relay selection method for optimizing transmission power between source node and relay is proposed.Use the proportion of transmission power between the source node and the relay to calculate the transmission power of the source node,select the relay link with the maximum channel capacity for data forwarding,and consider the transmission power constraints of the source node and the relay.On this basis,the average channel capacity and outage performance of the system are deduced.Simulation results show that the proposed method can significantly improve the energy efficiency of the system,and simultaneously save energy consumption of the source.
-
Computer Engineering. 2019, 45(9): 87-94. https://doi.org/10.19678/j.issn.1000-3428.0052263In order to improve the Quality of Service(QoS) of terminal in the area of dense mobile terminal business district,a network access method for dense mobile terminal network based on business-driven resource reservation is proposed.It reserves resources for terminals with high-priority business based on the attributes of networks,the forecasted motion trajectory of terminals and the types of business.Under the premise of minimizing the sacrifice of QoS of the low-priority service,it ensures high-priority business terminals get a smooth and reliable network service,and ensures network access resources for terminals with low-priority business by using time-triggered fairness mechanism.Simulation results show that,this method can not only ensure the access quality of mobile terminals with high-priority business in heterogeneous network and improve the utilization of network resources,but also reduce the access failure rate of low-priority business terminals in hot-spot high-capacity scenarios.What's more,it can improve the users'Quality of Experience(QoE).
-
Computer Engineering. 2019, 45(9): 95-99,118. https://doi.org/10.19678/j.issn.1000-3428.0051948The polling protocol is a classic Media Access Control(MAC) layer Time Division Multiple Access(TDMA) time slot allocation protocol,which has the characteristics of collision-free data communication and low power consumption,but it cannot guarantee communication reliability and realize differential services.Therefore,a TDMA time slot allocation protocol with dynamic retransmission and differential service mechanisms is studied.The frame structure is redesigned,and the static time slot allocation is completed according to the information of the terminal node data packet.When the data transmission fails,the retransmission time slot is used to improve the communication reliability,and the differential service is realized by the shadow time slot.Experimental results on the wireless meter reading system show that compared with the classic polling protocol,the packet loss rate of the protocol is reduced by 3%.
-
Computer Engineering. 2019, 45(9): 100-104. https://doi.org/10.19678/j.issn.1000-3428.0051618In order to improve the system energy efficiency of the Femto network,this paper proposes a power control algorithm supporting adaptive pricing.The adaptive pricing factor is defined as the edge drop of energy efficiency of the network when the home base station increases unit of power.Each home base station calculates an adaptive pricing factor based on the collected transmission power information of other home base stations,and distributes transmission power to the home base stations with energy efficiency first principle.Then,the power of each home base station is distributed using a distributed power control method.On this basis,the average energy efficiency of the system is obtained through multiple iterations.MATLAB is used to build the system simulation platform.The simulation results show that,compared with the power control algorithm based on fixed pricing,the proposed algorithm improves the system energy efficiency by about 18.6%.
-
Computer Engineering. 2019, 45(9): 105-111. https://doi.org/10.19678/j.issn.1000-3428.0053690In order to solve the network security defense problem after the intrusion detection system misses detection,using the Moving Target Defense(MTD) technology of the platform layer,this paper designs a new dynamic defense strategy for platforms.Combined with the dynamic defense system of the platform in Demilitarized Zone(DMZ),the dynamic defense operation principle of the platform is analyzed.Considering the active platform migration at fixed intervals caused by missed intrusion detection and random platform migration caused by detected intrusion alarms,a dynamic defense strategy for platform and an evaluation index for defense effects are given.Simulation results show that the proposed strategy always maintains a high cost-effectiveness ratio in the case of abnormal events and an increased false negative rate of the intrusion detection system,which can effectively solve the imbalance between defense costs and effects and the security risks caused by missed detection of the intrusion detection system.
-
Computer Engineering. 2019, 45(9): 112-118. https://doi.org/10.19678/j.issn.1000-3428.0051692Aiming at the Advanced Encryption Standard(AES) algorithm,this paper proposes a simple and efficient method of fault injection attack.By instantaneously reducing the voltage of the cipher chip power supply,the low voltage burr is generated,which causes the chip to skip the assignment cycle sentence when the key is assigned in the key expansion function,resulting in a key assignment error,thereby reducing the length of the unknown byte participating in the operation key.Combined with the error ciphertext output after the injection of the fault,the exhaustive guess can be use to recover the initial key unknown byte.Attack test results indicates that an effective fault injection attack can get a 4-byte length initial key.That is,for the 128-bit AES algorithm,the attacker can recover the initial key only by guessing 4×232 times.
-
Computer Engineering. 2019, 45(9): 119-123. https://doi.org/10.19678/j.issn.1000-3428.0051787To address the high complexity of the location algorithm based on the topology of the whole network,which is led by a coverage overlap between malicious information sources,a subgraph division algorithm based on community structure is proposed.The algorithm decomposes the location problem of multi-source malicious information into multiple single-source location problems.On this basis,the multi-source location algorithm for Online Social Network(OSN) based on the Jordan center is used for the single-source malicious information location in multiple subgraphs.Simulation results on the random number network and UCIonline network show that the algorithm can effectively identify malicious information sources,and its location accuracy is 11%~30% higher than that of algorithm based on Distance Center(DC),Tightness Center(TC) and Betweenness Center(BC).
-
Computer Engineering. 2019, 45(9): 124-127. https://doi.org/10.19678/j.issn.1000-3428.0050912The Akbary-Ghioca-Wang(AGW) criterion and piecewise method are two main methods for constructing permutation polynomials of finite fields.This paper introduces the application of permutation polynomials in cryptography and coding theory,reviews the research progress of the permutation polynomials and their inverses constructed by AGW criterion and piecewise method,describes the problem of permutation polynomials,and finally the next step is to look into the research work.
-
Computer Engineering. 2019, 45(9): 128-135,142. https://doi.org/10.19678/j.issn.1000-3428.0052317In order to accurately evaluate the vulnerability of computer network,a new evaluation algorithm is proposed by combining Bayesian network with attack graph.An attack graph model is constructed,which is named RSAG.On the basis of eliminating the loop in the attack graph,the model is transformed into a Bayesian network attack graph model,which is named BNAG,and the node accessibility probability is calculated by introducing the node attack difficulty and node state transition measurement index.The analysis results of an example show that the evaluation results of network vulnerability by this algorithm are true and effective,which can fully reflect the difference between attacked node.Meanwhile,the calculation of attack graph with mixed structure is less,which can accurately highlight the harm degree of vulnerability in the chaotic relationship.
-
Computer Engineering. 2019, 45(9): 136-142. https://doi.org/10.19678/j.issn.1000-3428.0051776In order to reduce energy consumption and improve data security and confidentiality in the process of data collection,processing and transmission in Wireless Sensor Networks(WSN),a dynamic clustering secure data aggregation algorithm CDSDA is proposed.The clustering-based WSN secure data aggregation algorithm CSDA is improved.According to the importance of data collected by nodes and the residual energy of nodes,the data is sliced dynamically.On this basis,intra-cluster aggregation and inter-cluster aggregation are carried out to get the final results.Simulation results show that,compared with CSDA algorithm,CDSDA algorithm has better performance in communication overhead,data accuracy,data privacy protection and so on.
-
Computer Engineering. 2019, 45(9): 143-146,152. https://doi.org/10.19678/j.issn.1000-3428.0053715A multi-bit homomorphic comparison operator supporting parallel acceleration is designed to achieve efficient comparison operation in dense state computing.A single-bit homomorphic digital comparator is constructed based on cuFHE software library,and it is called under parallel computing mode.A multi-bit homomorphic comparison operator that can compare plaintexts of any length is implemented by GPU hardware.The cuFHE homomorphic algorithm library is used to write homomorphic comparison operation function,and the function is tested.Results show that the comparison operator is more efficient,and only takes 0.91 s to perform a comparison operation on an 100-bit plaintext.
-
Computer Engineering. 2019, 45(9): 147-152. https://doi.org/10.19678/j.issn.1000-3428.0052363SASAKI Y,et al used message modification technology to differentiate attack on 51-step RIPEMD-160(Lecture Notes in Computer Science,Vol.7341),but they did not consider the influence of modular subtraction difference on differential routes,and could not guarantee that the probability of the validation of the first half of the left and right operations is 1,which led to the error in the complexity of RIPEMD-160.For this reason,RIPEMD-160 differentiator is constructed by 2-dimension sums method and an improved distinguishing attack complexity analysis method is proposed.Sufficient conditions are given to ensure the validation of modular subtraction difference in differential routes.The probability of the validation of the first half of differential routes of left and right operations is increased from 2-7.717 to 1 after message modification.The probability of the validation of differential routes of the second half is obtained through experimental tests.The analysis results show that the complexity of distinguishing attacks on 51-step RIPEMD-160 is 2152.672.
-
Computer Engineering. 2019, 45(9): 153-160,182. https://doi.org/10.19678/j.issn.1000-3428.0053945To address the problems of time-consuming bilinear pairings and the heavy burden of PKI certificate management,WANG Z,et al.proposed an identity-based efficient Authentication Key Agreement(AKA) protocol for mobile Internet(Journal of Communications,2017,No.8),but this protocol cannot resist the attack of temporary private key leakage and does not have eCK security.For this reason,an identity-based authentication AKA protocol without bilinear pairings operation is proposed,whose eCK security is proved under the GDH assumption and random oracle model.Analysis results show that this protocol only needs four elliptic curve point multiplication in the key agreement stage,which is more efficient than CKD-10,XW-12,WML-17 and other protocols.Meanwhile,its single round communication times and computational cost are less,which is suitable for mobile Internet environment.
-
Computer Engineering. 2019, 45(9): 161-168. https://doi.org/10.19678/j.issn.1000-3428.0051486A malicious domain name detection method based on Hidden Markov Model(HMM) is proposed.The characteristics of good and evil domain name in DNS communication are analyzed,and the attribute characteristics are counted by using the efficient computing power of Spark big data processing platform.On this basis,malicious domain name are accurately classified by Baum-Welch algorithm and Viterbi algorithm in HMM.Experimental results show that compared with the Random Forest(RF) model,the accuracy and recall rate of HMM for malicious domain name classification are both higher.
-
Computer Engineering. 2019, 45(9): 169-175. https://doi.org/10.19678/j.issn.1000-3428.0052234The spiking neuron supervised learning algorithm adjusts the synaptic weight of the neuron by gradient descent method,but the accuracy gets low and the learning period gets long as the length of the target learning sequence increases.Therefore,a gradient descent learning algorithm with delay adjustment is proposed.The delay of each synapse is used as a learning parameter to adjust the weight during the learning process,and at the same time,the gradient delay adjustment of the synaptic delay time is performed to make neurons trigger a target spiking sequence.Experimental results show that the proposed algorithm can improve the ability of neurons to learn complex spiking sequences without increasing the complexity of the algorithm,and the convergence speed is faster.
-
Computer Engineering. 2019, 45(9): 176-182. https://doi.org/10.19678/j.issn.1000-3428.0051339The traditional recommendation system relies on manual rule design and feature extraction,resulting in insufficient extraction of the features and implicit information of the review text content.Aiming at this problem,a recommendation method for deep modeling of review text is proposed combining the attention mechanism and the improved recommendation system based on deep learning.The word embedding model is used to express the semantics in dataset comments,the attention mechanism is introduced to re-weight the input content,and the hidden features in the user and project comment data are mined through the parallel convolutional neural network.The two sets of features are coupled and input and scored by a factorization machine to obtain the recommendation results.Experimental results show that the proposed algorithm can efficiently improve the recommendation precision,and the mean square error is improved by more than 2% than that of the DeepCoNN algorithm.
-
Computer Engineering. 2019, 45(9): 183-187,193. https://doi.org/10.19678/j.issn.1000-3428.0051013In order to improve the text classification accuracy of feature extraction method,combining with Information Gain (IG) and improved minimal Redundancy Maximal Relevance(mRMR),an IG_CDmRMR two-stage text feature selection method is proposed.The IG is used to extract the feature set with strong correlation with the category.The class difference degree is used to dynamically change the weight of the mutual information value between the feature and the category in the mRMR,and the optimal feature subset is filtered to obtain the text categorization result.Experimental results show that compared with the IG method and the IG_mRMR method,the accuracy of the proposed method is improved by about 2% with the same number of features.
-
Computer Engineering. 2019, 45(9): 188-193. https://doi.org/10.19678/j.issn.1000-3428.0052195Aiming at the problem that the Convolutional Neural Network(CNN) parameters are large and difficult to be transplanted to the embedded platform,a CNN compression method based on Lempel-Ziv-Welch(LZW) encoding is proposed.The model capacity is compressed by two ways:convert floating point to fixed point and pruning.The weights are quantized by k-means clustering,and LZW encoding is performed on this basis.Experimental results on the MNIST dataset show that the pruning effect is better than the compression effect of converting floating point to fixed point.After pruning and quantification,the compression ratio of LZW encoding can reach 25.338.
-
Computer Engineering. 2019, 45(9): 194-197,203. https://doi.org/10.19678/j.issn.1000-3428.0052044Dendritic Cell Algorithm(DCA) requires the input of 3 types of signals,which needs to extract features in advance through manual selection or statistics.To extract features accurately and efficiently,a DCA combined with XGBoost is proposed.The XGBoost algorithm is used to iteratively generate decision tree.According to the feature nodes of the decision tree,the characteristic indexes of the data set are extracted and classified,and used as the signal input of DCA to optimize the algorithm.Experiments are carried on KDD99 dataset,and the results show that the algorithm has higher accuracy than the improved algorithm based on rough set,which can reach up to 0.859 00.
-
Computer Engineering. 2019, 45(9): 198-203. https://doi.org/10.19678/j.issn.1000-3428.0052139To address Slater voting which is NP-hard problem,this paper proposes an Answer Set Programming (ASP) method for calculating voting result.The method constructs the minimum number of no circuit tournament,finds the one with the smallest difference from the original tournament,and selects the winner.Experimental results show that the method is coded in a way independent of the number of candidates.Also,the readable method has a low time complexity,and is suitable for Kemeny elections.
-
Computer Engineering. 2019, 45(9): 204-210,234. https://doi.org/10.19678/j.issn.1000-3428.0052381To address the poor performance of the Smooth Inverse Frequency(SIF) model in text classification and sentiment analysis,based on the SIF model,the Variance Weight(VW) of the task contribution is calculated according to the distribution of words in different classification task,and a VW factor selection sentence vector model CwVW-SIF is established.Tested on the standard text classification datasets and sentiment analysis datasets,the results show that CwVW-SIF has higher classification accuracy than SIF model.
-
Computer Engineering. 2019, 45(9): 211-215. https://doi.org/10.19678/j.issn.1000-3428.0053470Aiming at the problem that the pruning condition of the rough set classification rule mining algorithm LEM2 is too strict,this paper proposes an improved rule mining algorithm for weighted fuzzy rough sets.In the framework of fuzzy rough set theory with weights of use cases,the mining method of classification rules for mixed data is discussed.The approximate covering parameters of rough set model are introduced as the generalization metric parameters of mining algorithm,which realizes the adjustment of the number of rule sets and the complexity of the rule form.Experimental results show that compared with LEM2 and DataSqueezer algorithms,the average precision and average recall of the proposed algorithm are better,reaching 81% and 80% respectively,and the average length of the generation rule is the shortest.
-
Computer Engineering. 2019, 45(9): 216-221. https://doi.org/10.19678/j.issn.1000-3428.0053937To address the problem that the Siamese Region Proposal Network(RPN) is susceptible to interference and cannot be tracked after the target is lost,this paper introduces the anchor mask network mechanism to design a new Siamese RPN model.The model sets the multi-scale template images and convolves them with the target image to achieve full-image detection and avoid target loss.By learning the IOU hot maps of the first three frames,the target anchor mask is predicted in the continuous frames to simplify the calculation and exclude other target interference.The experimental results in the VOT2016 and OTB100 datasets show that the model has a detection rate of 24.6 frame/s and an expected average overlap of 0.344 5 for the VOT2016 dataset,and a precision of 0.862 and a success rate of 0.642 for the OTB100 dataset.The target loss and interference tests are carried out on the data collected by the camera.The results show that the model has good anti-interference and real-time performance.
-
Computer Engineering. 2019, 45(9): 222-234. https://doi.org/10.19678/j.issn.1000-3428.0051964The Generative Adversarial Networks(GAN) based on the zero-sum game idea can obtain the distribution of data through unsupervised learning and generate more realistic data.Based on the basic concepts and theoretical framework of the generated confrontation network,the GAN models and the application results in specific fields are studied,and the data similarity measure,model framework and training method are summarized.The research and related research results of the improvement and expansion are analyzed,and the practical application fields such as image synthesis and style migration are discussed.The advantages and disadvantages of GAN are summarized,and the application prospects are prospected.Analysis results show that the GAN has strong learning ability and plasticity,great potential for improvement and wide application range.However,its development challenges are unstable training process and lack of objective evaluation criteria for generating data quality.
-
Computer Engineering. 2019, 45(9): 235-241,247. https://doi.org/10.19678/j.issn.1000-3428.0051786In the image dehazing process,the estimation of the atmospheric light transmittance is inaccurate,which reduces the scene brightness and causes halo effects in the sky of the dehazed images.To this end,this paper proposes an image dehazing algorithm based on block optimization transmissivity and adaptive scene brightness optimization.This algorithm performs block optimization on transmissivity based on the foggy degree of the image,uses the intensity of airlight to solve the atmospheric scattering model to generate a dehazed image,and then adaptively performs partial adjustment to the gray value of the image to increase scene brightness.Experimental results show that compared with the guided image filtering algorithm and contrast enhancement algorithm,the image dehazed by the proposed algorithm is clearer.The edge-preserving effect is obvious,and the visual effect is better,which means the proposed algorithm is suitable for application fields such as traffic supervision,security monitoring and target recognition.
-
Computer Engineering. 2019, 45(9): 242-247. https://doi.org/10.19678/j.issn.1000-3428.0051855In order to study the image features that can effectively reflect visibility,and solve the difficulties to structure a large-scale training data set,this paper proposes a method for applying deep convolution neural networks to visibility detection.The sample image is divided into several subdomains and encoded by using the pre-trained VGG-16 network.Support Vector Regression(SVR) models are trained with coded feature sets,and each subdomains' fusion weight is calculated according to the error analysis of the support vector,and then fuse the visibility estimates of subdomains by weight.Experimental results show that the detection accuracy of the proposed method exceeds 90%,which can meet the requirements of application.
-
Computer Engineering. 2019, 45(9): 248-252. https://doi.org/10.19678/j.issn.1000-3428.0051158The RC algorithm introduces regional-level contrast and re-quantizes the color model,which can greatly improve processing speed and highlight salient targets.However,the graph-based segmentation algorithm is prone to have the problem that the segmented region cannot fit the edge of the object well.This paper introduces an optimized SLIC algorithm instead of a graph-based segmentation algorithm to improve the RC algorithm,and implements an automatic matting system based on image saliency recognition,which overcomes the shortcomings of traditional matting systems that must be manually labeled.Experimental results show that compared with other classical algorithms such as IT,MZ,GB,RC,etc.,the improved RC algorithm captures salient targets with greater accuracy,its Precision,Recall,and F value are 0.82,0.85,and 0.83,respectively.The system can automatically capture salient targets and provide image synthesis applications.
-
Computer Engineering. 2019, 45(9): 253-259. https://doi.org/10.19678/j.issn.1000-3428.0054281Complex and changeable underwater environment leads to the poor quality of sonar images,decreasing the accuracy of target recognition.Therefore,a level set sonar image segmentation algorithm based on multiscale Gaussian Markov Random Field(GMRF) model under Contourlet domain is proposed.Contourlet transform and inverse transform are used to obtain the texture feature under each scale layer of the sonar image.The texture feature of each layer is modeled by GMRF to describe the local structure spatial information and reduce the sensitivity to noise.Based on the texture feature models of each layer,coarse-to-fine segmentation for level sets is performed on sonar images to obtain segmentation results.Experimental results show that the accuracy of the algorithm exceeds 90% in different sonar images,which is better than Otsu algorithm and has lower complexity and stronger robustness.
-
Computer Engineering. 2019, 45(9): 260-263. https://doi.org/10.19678/j.issn.1000-3428.0050017Existing generation technologies for directional excitation and random excitation is time-consuming,and result in heavy workloads for cross-platform use.To address the problem,a universal configurable data exciter is designed based on internal logic automata.The exciter combines autonomous data excitation and responsive excitation.Results of simulation on the combat vehicle navigation equipment show that the exciter has high accuracy and efficiency.The average response time to an event is 0.493 ms and the average formation time of the logic automata is 0.13 ms.
-
Computer Engineering. 2019, 45(9): 264-269. https://doi.org/10.19678/j.issn.1000-3428.0051861Aiming at the problems of repeated switching of positioning mode and low positioning accuracy of indoor and outdoor GPS/WiFi signal interleaving area,a fusion positioning method based on grey prediction model is proposed.The GM(1,1) model is constructed to obtain the object's grey prediction trajectory under the current positioning mode.The precise positioning is realized by using geometric method to deeply fusion the predicted trajectory with the positioning information in the positioning mode to be switched.When the fusion between the predicted trajectory and positioning information fails,the number of satellite signals will be used to decide whether to switch the location mode.Experimental results show that compared with the GPS-based and WiFi-based positioning method,the proposed method has better positioning precision and robustness.
-
Computer Engineering. 2019, 45(9): 270-275. https://doi.org/10.19678/j.issn.1000-3428.0052458The traditional Hilbert-Huang Transform(HHT) has low accuracy in Empirical Mode Decomposition(EMD) when processing pulse signals,and there are mode mixing problems.In response to this problem,this paper proposes an improved HHT method.Combined with Time Varying Filtering(TVF),the pulse signal is decomposed by EMD to obtain a series of Intrinsic Mode Functions(IMFs).The correlation coefficient method is used to extract the effective IMF components,to which the Hilbert transform is applied to obtain the Hilbert spectrum and the marginal spectrum of the pulse signal.Experimental results show that the proposed method improves the accuracy of decomposition,effectively solves the mode mixing problem,and can remove the interference components in the signal.The obtained Hilbert spectrum and the marginal spectrum can more accurately reflect the time-frequency characteristics of the pulse signal.
-
Computer Engineering. 2019, 45(9): 276-283,290. https://doi.org/10.19678/j.issn.1000-3428.0052140In Question Answering(QA) system,the query intent that is expressed only by the words in the question may not be able to get the ideal answer from the data source.To this end,a semantic query expansion method is proposed for QA system based on knowledge graph.The query terms in question triples are expanded from three semantic aspects (i.e.synonym,hypernym and hyponym) based on WordNet and other two semantic aspects (i.e.hypernym and hyponym) based on Microsoft Concept Graph.Different filtering strategies are designed to select the expansion results for each semantic aspects.According to the results of the semantic expansion of query terms,the expansion of question triple is achieved.Experimental results show that the proposed method can effectively expand the question triple in multiple semantic aspects,whose Precision is more than 83%.
-
Computer Engineering. 2019, 45(9): 284-290. https://doi.org/10.19678/j.issn.1000-3428.0052098In the intelligent customer service question-answering system,the questions asked by users are characterized by complex consultation intention,weak contextual relevance,and serious colloquialization.As a result,the accuracy of question similarity calculation is not high and irrelevant answer occurs.In order to solve these problems,a similarity calculation model MA-CNN based on Convolutional Neural Network (CNN) is proposed.Focusing on the semantic information between words and the overall semantic information between sentences through two different attention mechanisms,the problem understanding ability of the intelligent customer service can be improved.Experimental results show that,compared with the models based on word vector and Recurrent Neural Network(RNN),the MA-CNN model has stronger ability to identify questions,and its F1 value can reach up to 0.501.
-
Computer Engineering. 2019, 45(9): 291-295,301. https://doi.org/10.19678/j.issn.1000-3428.0051708In order to improve the fault handling capability of the test crew in the aero-engine development process,a visual test-run semi-physical simulation system is designed to conduct fault simulation in the air start system,fuel system and engine system.A fault processing model combining fault mathematical model,phenomenon mathematical model and troubleshooting method is established to realize fault setting,phenomenon and process simulation,and key performance index for fault elimination operation is evaluated according to system response time and precision.Simulation results show that the system can realize the visualization of the aero-engine test-run failure process and can directly reflect the fault phenomenon and the process.
-
Computer Engineering. 2019, 45(9): 296-301. https://doi.org/10.19678/j.issn.1000-3428.0051794The traditional robot PID control method has poor robustness and poor stability in complex environment because of its simple structure,while the closed loop feedback mechanism of the traditional Visual Servo (VS) control system affects the control effect of strong coupling system.Aiming at the shortcomings of traditional robot PID control and VS in the tracking problem of rotor UAV,this paper proposes a decoupled VS tracking method.The feature extraction algorithm based on Fehrman chain code is used to extract feature points of target and the VS model of the error of feature point to the velocity is established.The influence of the pitch angle and the rolling angle is removed by the attitude adjustment,the motion parameters are extended from four to six dimensions by the method of dynamic expansion,and the linear velocity servo gain and angular velocity servo gain are set respectively.The tracking experiment of rotor UAV under simulation condition and physical condition is designed.Experimental results show that compared with the traditional VS method and PID method,the proposed method has better target loss probability and better tracking effect.
-
Computer Engineering. 2019, 45(9): 302-309,315. https://doi.org/10.19678/j.issn.1000-3428.0052196In order to understand the law of rumor propagation deeply,a rumor propagation model for crowd-gathered place of emergency event is proposed.According to the different motives of gathering individuals in emergencies,the gathering people are divided into six different states,and the individual's personality characteristics are reflected by the individual's trust,rumor acceptance and social bad cognition.At the same time,the model analyzes the influence of factors such as the proportion of troublemakers,personality characteristics of ordinary citizens,and ways of handling event on the process of rumor propagation.Experimental results show that the personality social cognition and event handling methods of the model can affect the speed of rumors,while the proportion of initial communicators and the social relationship of the gathering people have little effect on rumors.
-
Computer Engineering. 2019, 45(9): 310-315. https://doi.org/10.19678/j.issn.1000-3428.0052127Due to the variations of geological structure in different earthquake regions,earthquake prediction in the sub-regions after dividing may not be well performed by a simple model.To solve this problem,a hybrid algorithm based on region segmentation is presented.After region segmentation,Back Propagation Neural Network (BPNN) algorithm and Dendritic Cell Algorithm(DCA) are respectively used to model each sub-region,and the better one is selected as the prediction algorithm of the sub-region.The earthquake occurrence situation in the whole region is calculated according to the weights of each sub-region.Experimental results show that the proposed algorithm can improve the effect of earthquake prediction.
-
Computer Engineering. 2019, 45(9): 316-320. https://doi.org/10.19678/j.issn.1000-3428.0052280To address the link prediction problem based on structural similarity in complex networks,based on the comparision of the similarity indexes of the existing link prediction algorithms,this paper defines a similarity index LRPA.LRPA is based on the node resource sharing concept in the resource allocation algorithm,the relationship between node degree and joint edge probability in the preferential attachment algorithm,and the local path. Based on LRPA,a new link prediction algorithm is proposed.The predictions are performed on the classical complex network datasets and the real bitcoin OCT trading network,experimental results show that,the proposed algorithm can accurately predict the edge structure and the trading mode of the bitcoin users.
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