15 June 2018, Volume 44 Issue 6
-
Select all|
-
Computer Engineering. 2018, 44(6): 1-7. https://doi.org/10.19678/j.issn.1000-3428.0047069
Traditional Graphics Processing Unit(GPU) executing PGrid index K-Nearest Neighbor(KNN) query method has problems such as large query granularity,redundant calculation,and unstable performance.Therefore,facing spatial KNN relation queries,a KNN query strategy based on fine-grained partition search is proposed.Based on the Euclidean distance trigonometric inequality feature,the dynamic query range expansion based on Cell is constructed to realize the fine-grained partition and expansion of the query range with respect to each boundary distance of the Cell,and the optimum grid scales of object numbers for a given K value scale are theoretically analyzesd.Experimental results show that compared with the traditional KNN query method,the query strategy has obvious performance advantages under different K values and grid division scales.
-
Computer Engineering. 2018, 44(6): 8-12. https://doi.org/10.19678/j.issn.1000-3428.0048320
For the detection of abnormal electricity behavior by users,power companies usually adopt manual inspection methods,however,this method requires a lot of manpower and material resources,and is influened by subjective factors.Therefore,an abnormal electricity behavior recognition algorithm based on ridge regression model is proposed.By collecting user electric data,the model is trained and the trained model is used for automatic detection of abnormal electricity behavior.In order to capture the sample information of unknown users’ electricity behavior categories,the graph regularization term is introduced on the basis of the ridge regression model.Taking into account the non-linear distribution characteristics of electricity data,the original data is mapped to high-dimensional Hilbert spaces through kernel functions,and a nonlinear ridge regression model based on graph regularity is gained.Experimental results show that compared with least squares,ridge regression,and graph regularization ridge regression models,this algorithm has higher recognition rate.
-
Computer Engineering. 2018, 44(6): 13-17,23. https://doi.org/10.19678/j.issn.1000-3428.0047313
To solve the problems that some similar users will be missed and the basis for selecting nearest neighbors is single while user clustering,a recommendation algorithm based on clusterings of item rating and type rating is proposed.The user-item rating matrix and the user-item type rating matrix according to user’s rating records are firstly generated.The fuzzy C-means clustering is carried out by using the above two matrices and the improved the distance measurement method.Then,the nearest neighbor is selected according to the membership degree matrix generated by the clustering.Finally,the prediction rating is generated by the parameter weighting.Experimental results on MovieLens dataset show that the proposed algorithm can reflect the user’s rating accurately and improve the accuracy of the recommendation system effectively.
-
Computer Engineering. 2018, 44(6): 18-23. https://doi.org/10.19678/j.issn.1000-3428.0047207
Aiming at the problem that the accuracy of the multi-dimensional continuous data is too low for traditional naive Bayesian classification algorithm,an improved classification algorithm based on attribute association is proposed.Directed against the multidimensional continuous data set with different attribute classes,it discretizes the data set by Gaussian segmentation,which is improved by using Laplace calibration,attribute association and weighted attribute.Experimental results show that,compared with improved algorithms by Laplace calibration or attribute weighting,the proposed algorithm can improve the accuracy of classification results,and its amplitude increase is increased with the increase of the number of attributes in a certain range,which is suitable for the classification of multidimensional continuous data.
-
Computer Engineering. 2018, 44(6): 24-28. https://doi.org/10.19678/j.issn.1000-3428.0047104
In order to solve the problem of low connection efficiency after the Hash distribution table is converted into a random distribution table,a parallel connection operation algorithm for Hash tables in Massively Parallel Processor(MPP) database is presented.According to the data block distribution characteristics of the hash distribution table in the shared storage environment,combining with the scanning advantage of random reading,data multiple copies distributed storage is used to improve the local reading rate without losing the characteristics of data hash distribution.TPC-H standard test results show that compared with the traditional parallel connection algorithm,this algorithm can effectively improve the connection operation efficiency and reduce the response time of the connection query up to 30%.
-
Computer Engineering. 2018, 44(6): 29-33. https://doi.org/10.19678/j.issn.1000-3428.0047129
Cooperative spectrum sensing algorithm requires a control channel when detecting the spectrum of primary user.However,the limitation of control channel bandwidth is the bottleneck of application of cooperative spectrum sensing when the number of cognitive users becomes very large.In order to solve this problem,this paper proposes a multi-threshold cooperative spectrum detection algorithm.Based on the establishment of a spectrum detection model,cognitive users are clustered with different application weight requirements to avoid congestion in the control channel.Results of theoretical analysis and simulation on the Matlab platform show that,the proposed algorithm can reduce the network delay and improve the detection probability of the system in Rayleigh channels.
-
Computer Engineering. 2018, 44(6): 34-39. https://doi.org/10.19678/j.issn.1000-3428.0046980
It is difficult to maintain the connectivity when Primary User(PU) is active in Cognitive Radio Network(CRN).To solve the above problem,this paper presents an efficient scheme to build a bi-channel connected network with the minimum number of required channels combining power control and channel assignment.Firstly,basic topology is generated by using graph coloring to assign channels to every secondary user to achieve bi-channel connectivity.Secondly,considering local spanning tree is not connected after deleting nodes,it applies the improved MPH algorithm to give priority to link on these nodes which have the large path weights,which means they are passing by shortest paths.Then,aiming at the problem that the topology is divided into two parts after deleting the nodes,it adds node at the middle of the shortest link between all these edges,which can connect the two parts.Theoretical analysis and simulation results show that,the proposed scheme can maintain the connectivity of network,meanwhile reducing the mumber of required channels and network costs.
-
Computer Engineering. 2018, 44(6): 40-44. https://doi.org/10.19678/j.issn.1000-3428.0047435
As a novel Non-Orthogonal Multiple Access(NOMA) technology,Sparse Code Multiple Access(SCMA) can meet the performance requirements of massive connections for 5G,but the Message Passing Algorithm(MPA) has the problems of slow convergence and high complexity.In this paper,a multi-user detection algorithm for SCMA system with low complexity is proposed to slove above problems.By narrowing down the search range of superposed codeword constellation points and introducing the weighting factor,it changes the initial probability of superpoed codeword constellation points in the search range and makes the decoding process faster and more accurate.Theoretical analysis and simulation results show that the proposed algorithm not only reduces the complexity,but also accelerates the convergence speed of the iterative process.
-
Computer Engineering. 2018, 44(6): 45-49. https://doi.org/10.19678/j.issn.1000-3428.0046881
In cognitive radio network,spectrum sensing performance tends to be reflected by the system throughput.Based on the traditional perceptual frame structure,this paper redefines a new perceptual frame structure by introducing the cooperative spectrum prediction and spectrum segmentation.By combining the Hidden Markov Model(HMM) cooperative spectrum prediction algorithm based on DBSCAN,the accuracy of spectrum prediction is improved,the consumption of cooperative prediction bandwidth is reduced,and the effect of improving system throughput is achieved.Simulation results show that the system throughput under the improved frame structure is improved compared with that of the frame structure without the cooperation module and the frame structure with the cooperation module but without spectrum segmentation.
-
Computer Engineering. 2018, 44(6): 50-56. https://doi.org/10.19678/j.issn.1000-3428.0047190
There are problems of waveform distortion,instability and even losing efficacy on condition that the beamforming algorithm of traditional two dimensional angular domain is applied to Three Dimensional Multiple Input Multiple Output(3D MIMO) scene.To solve the problem,on the basis of the traditional Two Dimensional Minimum Mean Square Error(2D MMSE) algorithm,an improved 3D MMSE angle domain beamforming algorithm is proposed,which applies the method of the eigen-decomposition to decompose the signal correlation matrix,removes the fluctuating factors in small eigenvalues associated with noise,and solves the issue of waveform distorting and losing efficacy.Simulation results show that the proposed algorithm can implement the beam forming in 3D angular domain,so that the MSE of array output is smaller and the Signal to Interference and Noise Ratio(SINR) of array output is higher than the 3D MVDR algorithm and the ZF algorithm.
-
Computer Engineering. 2018, 44(6): 57-61. https://doi.org/10.19678/j.issn.1000-3428.0046907
Based on the inherent high sidelobe problem in Orthogonal Frequency-Devision Multiplexing (OFDM) system,a sidelobe suppression method for OFDM system is proposed based on subcarrier weighting method.The subcarrier weights are determined in such a way that the sidelobes of the transmission signal are minimized according to an optimization algorithm which allows several optimization constraints.To more effectively suppress the out-of-band radiation,the time domain windowing is introduced on the basis of the subcarrier weighting.The emission signal is multiplied by a window function which is different from the conventional rectangular window.Simulation results show that,the subcarrier weighting method can suppress the radiation intensity on average not less than 10 dB of OFDM sidelobe in the average without requiring the transmission of any side information,after the weighted transmission signal is windowed,the suppression effect of the sidelobe of the OFDM signal can reach about 25 dB.
-
Computer Engineering. 2018, 44(6): 62-67. https://doi.org/10.19678/j.issn.1000-3428.0046402
The nodes importance ranking of complex networks is very important for network survivability research.Most of the existing nodes importance ranking methods do not consider the influence of network structure change and important neighbor nodes.To solve this problem,on the basis of multi-attribute evaluation and nodes deletion,an improved ranking method for complex network nodes is proposed.According to the local network nodes attributes,global network properties and network location attribute,the evaluation index is selected,and the initial nodes important degree is evaluated by the ideal point method.Followed by deleting the most important node,important degree of remaining nodes are evaluated,the influence of network structure and important neighbor nodes are reduced in this way,and the important network nodes ranking is finally get.Experimental results show that,compared with other methods,such as ideal point method,NICCM method,the proposed method is more accurate for the recognition and ranking of important nodes.
-
Computer Engineering. 2018, 44(6): 68-73. https://doi.org/10.19678/j.issn.1000-3428.0046846
In Wireless Sensor Network(WSN),the current Energy-efficient Reliable Opportunistic Routing(EROR) algorithm based on network coding has shortcoming in balancing the energy consumption of nodes and prolonging the lifetime of network,so an optimized opportunistic routing algorithm based on network coding named OPEROR is proposed.It uses the channel bit error rate and packet loss rate to calculate the probability of failure of the received encode packets,reducing the retransmitted number of encode packets.The nodes in the forwarding set broadcast the cost packet directly instead of acknowledgement packet when it collects enough encode packets,reducing the network overhead and the waiting time of primary forwarding nodes.The neighbors of the primary forwarding nodes update their own forwarding cost based on the received cost packet,and decide whether to become assistant forwarding node,thus can prevent the neighbors with large forwarding cost from forwarding encode packets.Simulation results show that compared with EROR algorithm,the performance of OPEROR algorithm is improved in prolonging the lifetime of network and reducing the average energy consumption.
-
Computer Engineering. 2018, 44(6): 74-79. https://doi.org/10.19678/j.issn.1000-3428.0046972
The Low Duty Cycle-Wireless Sensor Network(LDC-WSN) greatly extends the lifetime of the network by making the node at a low duty cycle,but makes the sleep delay in the network longer.To solve this problem,a Sleep Scheduling algorithm based on Multiple Wake-up(SSMW) is proposed,which balances energy consumption and time.Through dynamic sensing of residual energy and the multiple wake-up mechanism,a lower bound of delay for any topological structure is given.Simulation results show that,compared with LES algorithm and TOSS algorithm,the proposed algorithm has a significant performance improvement in the sleep delay,which can balance the energy consumption and prolong the network lifetime effectively.
-
Computer Engineering. 2018, 44(6): 80-85. https://doi.org/10.19678/j.issn.1000-3428.0046325
In order to solve the hot pot problem which is caused by uneven load energy of cluster head node in Wireless Sensor Network(WSN),this paper proposes a partition routing protocol based on multiple sinks.It deploys multiple sinks and makes a reasonable partition in monitoring area,and uses the bat optimization algorithm which leads into variable scale chaos strategy to select the appropriate cluster head nodes to avoid the local optimization of bat algorithm.In data transmission phase,the cluster head node chooses the remaining energy greater than the average energy and the nearest node from the base station as the next hop.Simulation results show that compared with LEACH protocol and DEBUC protocol,the proposed protocol can reduce and balance energy consumption,increase the amount of packet reception,and extend the network lifetime.
-
Computer Engineering. 2018, 44(6): 86-92. https://doi.org/10.19678/j.issn.1000-3428.0046512
Faster-than-Nyquist(FTN) signaling,which is promising in 5G mobile communication,has higher information rate and spectrum efficiency.However,symbol interval’s compression resulted in infinite Inter-Symbol Interferences (ISI),which would deteriorate the performance of carrier synchronization.To solve the problem,the new carrier synchronization algorithm is proposed.The frequency can be obtained by the spectrum peak index,which can be obtained by the quadratic polynomial fitting for periodgram main lobe.In addition,the algorithm is improved by increasing the length of FFT,combining the interpolation and fitting algorithm.Simulation results indicate that improved algorithm can adapt to the FTN signaling character better than Candan’s algorithm and the adjacent symbol phase division algorithm.And the estimation performance can approach to Cramer-Rao Bound (CRB) better.
-
Computer Engineering. 2018, 44(6): 93-99,103. https://doi.org/10.19678/j.issn.1000-3428.0046548
Bandwidth utilization is a core index in network transmission capability evaluation.Too large bandwidth utilization will cause node congestion and make the performance of the network drop dramatically.Aiming at this problem,based on bandwidth utilization,a bandwidth utilization routing algorithm with exponential function is proposed,named EBURM.It firstly puts forward an objective function in mathematics,then studies and calculated several key elements and factors in the objective function,and gives the the theoretical values and calculation formula of transmission of efficiency,bandwith utilzation and step factor.In the current popular enterprise intranet network architecture,EBURM is compared with classical OSPF routing mechanism by using simulation tools.Experimental result shows that the EBURM has excellent linear characteristics under the condition and the number of EBURM path is less than five.EBURM can adjust the traffic transmission strategy of the source node in the network to reduce the cost in the traffic transmission.
-
Computer Engineering. 2018, 44(6): 100-103. https://doi.org/10.19678/j.issn.1000-3428.0048913Most of existing algorithms for path coverage do not consider minimization of sensors movement distance.Aiming at this situation,this paper proves that the above mentioned problem is NP-hard,and presents a heuristic path covering algorithm.It moves sensors through techniques of path curve discretization,redundant sensors and redundant paths,resulting in coverage of the whole path with minimum movement distance of sensors.It gets analysis results that,when m sensors cover n points on the path,the time complexity of the algorithm is O(n4m+n3m2).Simulation experiments illustrate that the proposed algorithm can reduce time complexity and moving distance when the number of path points and system parameters change.
-
Computer Engineering. 2018, 44(6): 104-110. https://doi.org/10.19678/j.issn.1000-3428.0047884Aiming at the problem of network security caused by ransomware,based on the analysis of a large number of ransomware samples,a method of detecting and analyzing ransomware based on dynamic symbol execution is proposed.A ADRAS system model is built based on piling tool Pin and constraint solver STP,by using dynamic symbol execution and Satisfiability Modulo Theories(SMT),the ADRAS system model can monitor the encryption function of the ransomware,and effectively capture ransomware’s encryption behavior and related information,and finally detect the ransomware of a number of families.Experimental results show that,the ADRAS system model can detect 15 samples from known ransomware families,including the famous CryptoLocker and the recent outbreak of WannaCry.
-
Computer Engineering. 2018, 44(6): 111-116. https://doi.org/10.19678/j.issn.1000-3428.0049841
In order to improve the efficiency and security of data upload for cloud storage users,a cloud storage channel hiding algorithm based on multiple bit repeating data deletion is proposed.A framework for cross user covert channels including cloud storage providers,victims and attackers is built,and the use of the message selection upload mechanism eliminates the need for “0” file upload,so as to reduce the number of uploaded files.At the same time,in order to realize error free decoding of multiple bit covert channel,a new synchronization technique is proposed,which sorts files with time stamp,so as to improve the order of data transmission.Finally,in the cloud storage service SugarSync and BaiduYun,the performance of the proposed algorithm is tested,and the results verify the effectiveness and security of the algorithm.
-
Computer Engineering. 2018, 44(6): 117-121,129. https://doi.org/10.19678/j.issn.1000-3428.0048197The leak of Application Programming Interface(API) key may cause the illegal use of services,and then lead to unpredictable economic losses.The common characteristics of API keys in different project codes are extracted by analyzing the basic characteristics statistics and the source code static structure of the samples.Then,an automatic identification method based on machine learning is built to detect the API keys in the source code.The result of 10-fold cross-validation experiment results show that the identification method is better in retrieval performance than traditional detection approaches such as full-text matching search,keywords search and information entropy search.
-
Computer Engineering. 2018, 44(6): 122-129. https://doi.org/10.19678/j.issn.1000-3428.0047241Aiming at the complexity and high frequency of fraud risk existing in the current financial payment industry transactions,a new method called the Density Distribution Evolution-based Stream(DDE-Stream) clustering algorithm is proposed.The CLIQUE algorithm is used to cluster the trading features of the account,and an account trading behavior archive model is constructed by combining with hidden Markov algorithm.According to the model,real-time fraud detection can be performed.In the stage of model self-renewal,DDE-Stream algorithm is used to perform effective real-time clustering of multi-dimensional trading features.Experimental results show that the algorithm can detect the risk of transaction fraud in real time,and the accuracy rate obtained on the verification set exceeds 50%,compared with the traditional random forest classification algorithm.
-
Computer Engineering. 2018, 44(6): 130-135,140. https://doi.org/10.19678/j.issn.1000-3428.0046985
In order to study the influence of different edge immunization strategies on the spread of rumors and synthesize the effects of different types of triple group structure on the nodes,a new edge immunization strategy,called triple group immunization strategy is proposed.Simulations are performed on real online social networks and artificially synthesized scale-free networks,respectively,and degree-degree edge immunization,mediation-mediacy edge immunization,feature vector-feature vector edge immunization,edge inversion immunization and triple group edge immunization equalization immune strategies are compared.Simulation results show that when the probability of infection of the rumors is low,the effect of the triple edge immunization is not significant,when the infection probability is high,the triple groupd edge immunization effect is better,and the effect is second only to degree-degree edge immunization.
-
Computer Engineering. 2018, 44(6): 136-140. https://doi.org/10.19678/j.issn.1000-3428.0046392
The two-dimensional code is only suitable for the transmission of single privilege information,cannot meet the needs of different permission users to gain the information.Using hierarchical encryption mode,the two-dimensional code information is encrypted according to the permission of secret information.The Hash function is used to generate the private key corresponding to different permission secret information automatically,to meet the acquisition needs of low permission information by high permission users.Using the attribute encryption algorithm,the access tree is generated according to the attribute set corresponding to the permission and access privilege of different information.The access permissions corresponding to different user attributes is calculated and the user private key is generated.The privilege of secret information and user attribute permissions are matched to complete the hierarchical encryption of two- dimensional code.Experimental results show that two dimensional code classification encryption algorithm based on attribute encryption can meet the needs of different permission users for different privilege information.
-
Computer Engineering. 2018, 44(6): 141-146,161. https://doi.org/10.19678/j.issn.1000-3428.0047079
To solve the problem of Electronic Control Unit(ECU) in the Controller Area Network(CAN) bus of intra-vehicle network,which is easy to be tampered and faked,this paper puts forward an authentication protocol based on one-time pad in the vehicle network.Firstly,it uses secure storage module TA in the Gateway ECU(GECU) to verify the legal identity of the ECU.Then according to the vehicle power supply voltage,it achieves the random number,generates the session key,and simplifies the key management of TA to ECU.At last,it periodically updates the session key when connecting and releasing external devices and effectively prevents the replay attack.The experimental result shows that this protocol can be applied to the vehicle environment efficiently,which significantly reduces the bus load and improves the communication efficiency.
-
Computer Engineering. 2018, 44(6): 147-150. https://doi.org/10.19678/j.issn.1000-3428.0046691
According to previous studies,the selection of the number of hidden nodes in the 5-levels neural network structure is not clear.To solve the problem,an improved 5-levels Deep Belief Networks(DBN) structure design and optimization method of nodes number is proposed.The number of the first hidden layer and the second hidden layer nodes is estimated to be 1/3 to 2/3 between the number of the first layer nodes.The number of the third hidden layer and the fourth hidden layer nodes equals the first of the number of hidden layer and input layer nodes,and then the first hidden layer and the second hidden layer nodes number is optimized by spline interpolation method.The structure features only 2-layers of weight before pre-training,which simplifies the Restricted Boltzmann Machine(RBM) pre-training method of DBN.Experimental results on the MNINST dataset verify the efficiency and high accuracy of the network structure.
-
Computer Engineering. 2018, 44(6): 151-155. https://doi.org/10.19678/j.issn.1000-3428.0047061On the study of lexical similarity,sememe similarity and concept similarity,this paper propose a sememe similarity computation algorithm based on the How-net semantic tree.This algorithm considers the node distance,the node depth and the number of brother node,so that gives out a lexical semantic similarity algorithm based on sememe similarity.Experimental results show that this algorithm can increase the accuracy of word semantic similarity and do not increase the complexity of algorithm compared with the tendency algorithm and semantics similarity algorithm in literature.
-
Computer Engineering. 2018, 44(6): 156-161. https://doi.org/10.19678/j.issn.1000-3428.0047342In order to solve the problem of data sparsity and cold start in the process of user recommendation,the similarity weighted recommendation algorithm is designed by constructing fuzzy time series classification model.A fuzzy time series classification model is established through combining user attribute label and time dimension while preprocessing data,and the blank data is predicted and filled by Lagrange interpolation.At the same time,aiming at the score difference caused by different users’ personal score too high or too low and unilateral rating problem,accuracy of the algorithm is promoted by similarity weighted fusion method.Experimental results show that the proposed algorithm can effectively reduce the Mean Absolute Error(MAE) and improves the recommendation quality.
-
Computer Engineering. 2018, 44(6): 162-168,175. https://doi.org/10.19678/j.issn.1000-3428.0047454The Coordination with Overt Conjunctions(COCs)in the Chinese patent literature are complex and widely distributed.The existing recognition technology can only use limited features to identify some simple types of parallel structures,and the recognition results are not very good as a whole.A method based on boundary-perceiving principles for recognizing COCs is introduced.Under the guidance of the HNC theory,COCs are annotated in the eight aspects:number,level,semantic type,semantic feature,interference,structural feature,contextual words and boundary position.The semantic characteristics,structural characteristics and contextual information are investigated and summarized;and 217 formal rules are set up and integrated into a HNC translation system.In contrast to Google Translate,the open experiment shows that this new method has better accuracy rate.
-
Computer Engineering. 2018, 44(6): 169-175. https://doi.org/10.19678/j.issn.1000-3428.0051068The dynamic summarization is to construct evolutionary content of collection.But there are some complicated problems in dynamic summarization,such as information redundancy,novelty information easily lost.To solve the above problems,this paper proposes an Integer Linear Programming(ILP) dynamic summarization update method based on Topic Signature model.According to the similarities between sentences,it calculates the representativeness score and diversity score for each sentence and introduces the Topic Signature model to determine the novelty of the sentences.Based on the summary generation strategy,the feasible region of understanding can be reduced and the high quality abstracts can be generated in a short time.Experiment result shows that the proposed method can effectively reduce the time complexity and improve the efficiency without model training and language matching.
-
Computer Engineering. 2018, 44(6): 176-181. https://doi.org/10.19678/j.issn.1000-3428.0047321In complex environments,the accuracies of traditional object detection algorithms is too low.A new pallet detection algorithm is proposed.A large number of pictures containing people and pallets in real warehouses are collected and annotated to building the pallets dataset for training.In the paper,the based network of the improved algorithm based on Single Shot MultiBox Detector(SSD) pipeline is adapted with DenseNet.In the experiments,the real pallets dataset is used for training and testing.In the testing phase,multi-scale feature maps of different resolutions are utilized for enhancing the network’s ability to adapt to detected objects,and the detection task is employing with single network.The experiment results show that the accuracy of the improved algorithm increased by 6.1% compared with YOLO algorithm.
-
Computer Engineering. 2018, 44(6): 182-187. https://doi.org/10.19678/j.issn.1000-3428.0047182Loop closure detection methods based on deep learning can achieve good detection performance in complicate illumination environment.But the dimension of extracted scene feature is too high to achieve the real-time detection requirement for closed loop.For that,propose a loop closure detection method based on fast and lightweight convolutional neural network.Concatenated rectified linear unit,batch normalization and deep residual module are combined to design the fast and lightweight network model which is trained with the Places365-Standard data set.The scene feature is extracted by trained neural network model and the regions of closed loop are obtained by measuring the similarities of them.Results from the test show that compared with loop closure detection algorithms based on PlaceCNN and Autoencoder,the proposed method not only achieves the high correct rate but also increases the implementation speed for detection.
-
Computer Engineering. 2018, 44(6): 188-193. https://doi.org/10.19678/j.issn.1000-3428.0047084In the existing human body behavior recognition algorithm for 3D space-time cubes,there is a problem that the time length is not accurately divided and the complexity of the dimension reduction processing time is high after feature extraction.Therefore,a 3D cube sparse coding recognition algorithm based on the trajectory of key points of human motion is proposed.Define the human body’s atomic behavior,the whole human body as a particle to get the particle trajectory,and according to the characteristics of the body’s atomic behavior on the trajectory segmentation,each trajectory represents an atomic behavior.The 3D space-time cube obtained by dividing the length of the trajectory segment as the segmentation length is divided by the atomic behavior time length.The key points in each frame of the cube are matched,and the key point trajectory is established.According to the characteristics of the trajectory,the dimension of the trajectory matrix is reduced by sorting,which effectively avoids complex matrix operations and reduces time complexity.Sparse coding of the dimensionality-reduced trajectory matrix is done,sparse expression under different motion modes is obtained.Experimental results show that compared with Blank,Castrodad and other algorithms,the algorithm has higher recognition rate and lower time complexity.
-
Computer Engineering. 2018, 44(6): 194-199. https://doi.org/10.19678/j.issn.1000-3428.0047276Aiming at the influence of occlusion,pose change,illumination change,complex background and other issues during the tracking process,an improved robust tracking algorithm is proposed.The sparse coding method of local blocks is used to combine the SIFT features and spatial pyramid feature matching techniques to study the objective modeling method.In the framework of particle filtering,motion estimation is introduced to obtain the optimal target position,and the template is updated in real time during the tracking process.Experimental results show that compared with common tracking algorithms such as DFT,FCT and OAB,the algorithm can still maintain good tracking robustness under complex conditions.
-
Computer Engineering. 2018, 44(6): 200-206. https://doi.org/10.19678/j.issn.1000-3428.0047330Aiming at the explicit problem of ignoring the characteristic words in the Latent Dirichlet Allocation (LDA) model,a topic model WL_LDA embedded tag information is proposed.A method for constraining unidirectional LBP images based on SIFT feature points is designed to obtain the texture structure of the image and the visual word is marked.By embedding the tag information in the LDA,the topic distribution of the image is derived using the two-dimensional histograms of words and tags.This topic distribution is used to train the classifier to complete the state detection of railway fasteners.Experimental results show that,compared with the LDA topic model,the differentiation degree of various types of fasteners in the topic space increases by 4.5% to 15%.Compared with the existing methods such as PCA and DF,the rate of missing detection and false detection is significantly reduced,it has better classification performance.
-
Computer Engineering. 2018, 44(6): 207-212,218. https://doi.org/10.19678/j.issn.1000-3428.0046520In the automatic reassembly of cultural fragments,because of the loss of the geometric characteristics caused by the damage of the fractured faces,an automatic reassembly algorithm based on the two-way distance field of the break-curves is proposed.Through the extraction of apparent ridges of fragments surface,so as to get the sculpture structure,using the minimum approximation error method to extract feature points on the contour surface decoration,on the basis of the vertex curvature value acquiring feature points of fractured faces,construction of bidirectional distance field fracture profile feature points to the surface and fracture surface feature points,the introduction of Euclidean distance from the consistency and complementarity of convex constraints,structuring feature descriptors,defining matching degree function to obtain the feature points matching set,and using quaternions algorithm to calculate the rigid transform matrix,the iterative closest point method to achieve accurate registration.Experimental results show that compared with the traditional splicing algorithm,the algorithm can save reassembly time and reduce the errors.
-
Computer Engineering. 2018, 44(6): 213-218. https://doi.org/10.19678/j.issn.1000-3428.0046638Aiming at the problem of accurate segmentation of overlapping pests in the counting process,an improved algorithm of segmentation of pests with pits detection and dividing point location is proposed.The overlapping area of pests is extracted by the shape factor and the pixel area factor constraint.The improved Harris algorithm is used to calculate the pixel similarity of the corner points to find the candidate corners.The non-pits in the candidate corners are removed and the remaining pits perform local non-maximal suppression to find true pits.The separation point of the overlapping pests is stripped layer by layer to find the separation points,and the final dividing points can be determined by the factor of pest pixel area constraint and the distance between the separated points and the pits.Experimental results show that the proposed algorithm can accurately find the dividing line of overlapping pests and make the operation more efficient.
-
Computer Engineering. 2018, 44(6): 219-225,232. https://doi.org/10.19678/j.issn.1000-3428.0048300
Traditional super-pixels segmentation method has problems of poor anti-noise performance,inaccurate merging and other issues.To solve these problems,a method of segmentation of side-scan sonar images by super-pixel clustering is proposed.The fast bilateral filter is used to perform the noise reduction processing to reduce the difficulty of subsequent segmentation.The luminance and texture features of the de-noised side-scan sonar images are extracted,and the similarities of the two features are calculated.These similarities are combined with weights to give the distance metric between the pixels and the cluster centers to generate the super-pixels.The saliency super-pixels are labeled based on the luminance feature,and all super-pixels are clustered by the maximum flow and minimum cut method.The proportions of the super-pixels with saliency within the clusters are calculated,and they are compared with the preset threshold.Clusters with proportion which is larger than the threshold are marked as foreground;otherwise,they are marked as background.As a result,the final segmentation is obtained.Experimental results show that compared with Fuzzy Local Information C-Means (FLICM) algorithm and Simple Linear Iterative Clustering (SLIC) algorithm,the segmentation accuracy of the proposed algorithm is high,and over-division rate and under-division rate are low.
-
Computer Engineering. 2018, 44(6): 226-232. https://doi.org/10.19678/j.issn.1000-3428.0046669When Adjacent Propagation(AP) clustering algorithm is used to segment color images,the similarity matrix has the problems of large scale,long clustering time,and high space complexity.Therefore,a new color image segmentation algorithm is proposed.A Simple Linear Iterative Clustering(SLIC) is used to pre-segment the color image,the L,a and b color component averages of each superpixel are calculated,and the similarity matrix of the AP clustering algorithm is constructed according to the negative Euclidean distance between the color component averages.An improved clustering method with decreasing reference degree is proposed during the AP clustering iteration process to improve the accuracy and robustness of AP clustering algorithm.Super-pixel clustering is used to evaluate the contour coefficients to obtain the best AP clustering results.Experimental results show that compared with the existing color image clustering and segmentation methods,the segmentation effect and segmentation quality of the algorithm are significantly improved.
-
Computer Engineering. 2018, 44(6): 233-238,248. https://doi.org/10.19678/j.issn.1000-3428.0047126In the moving object detection methods base on low rank and sparse decomposition,the nuclear norm is not the best approximation of the rank function of the matrix,meanwhile,the spatial continuity of moving object is not to be considered.As a result,the effect of moving object detection isn’t ideal on the circumstance of dynamic interference.To deal with this problem,a model named weighted truncated p-norm model is proposed.This model divides the observed video into three parts:the static background、the moving object and the dynamic background.The static background is constrained by an improved non-convex norm——weighted truncated p-norm.And according to the characteristic of spatial continuity,the dynamic background and moving object are constrained by structured-sparse l2,1-norm,respectively.Experimental results show that the WTPR model can remove noise of dynamic background effectively,also the more accurate moving object can be obtained than Robust Principal Component Analysis(RPCA)、Truncated Nuclear Norm Regularization(TNNR)、Weighted Nuclear Norm Minimization(WNNM)、Detecting Contiguous Outliers in the Low-rank Representation(DECOLOR).
-
Computer Engineering. 2018, 44(6): 239-248. https://doi.org/10.19678/j.issn.1000-3428.0046943The image quality-aware perception feature extraction and image quality value mapping are the core components of the Stereoscopic Image Quality Assessment(SIQA) algorithm.Related to 2D Image Quality Assessment(2D-IQA) algorithm,various factors affecting stereoscopic quality perception such as parallax,stereoscopic viewing comfort and stereo saliency need to be introduced in the image quality-aware feature extracted by SIQA algorithm,but there is little difference in image quality value mapping.Therefore,considering the parallax information as the main classification standard,the SIQA algorithm is used to extract various methods which the quality-aware features of stereoscopic images are introduced,and their characteristics are reviewed.The current problems and future research directions for extracting stereoscopic image quality-aware features are analyzed and forecasted.
-
Computer Engineering. 2018, 44(6): 249-252,258. https://doi.org/10.19678/j.issn.1000-3428.0046599During the pharmaceutical process,manual inspection suffers from time consuming and inefficiency.Image processing method needs plenty of samples and has poor adaptability.To address these problems,a method based on feature extraction and machine learning is developed.The Bag of Feature(BOF) algorithm is introduced to extract the features of the target region and its optimum parameter value is analyzed.Support Vector Machine(SVM) is applied to inspect the defects of the hierarchical drug.Experimental results show that the proposed method can achieve higher classification accuracy and meet the requirements of real-time inspection in the process of industrial production.
-
Computer Engineering. 2018, 44(6): 253-258. https://doi.org/10.19678/j.issn.1000-3428.0047348In order to solve the problem of low recall rate in object detection with the deep reinforcement learning method,on the basis of simulating human visual mechanism,a dynamic searching hierarchical offset method is proposed.It uses the idea of anchors based on the original hierarchical searching method,which adds a region offset.This method avoids the limitations generated by hierarchical searching method,and makes the search more flexible.This paper combines the advantages of Double DQN and Dueling DQN,using Double Dueling DQN network structure as the deep reinforcement learning network of the agent.Experimental results show that the accuracy and recall ratio are higher than the original hierarchical searching method.
-
Computer Engineering. 2018, 44(6): 259-262,269. https://doi.org/10.19678/j.issn.1000-3428.0048455In the traditional global stereo matching methods,pixel-based belief propagation has a large amount of computation and a single pixel point can easily cause errors.Therefore,on the basis of the image segmentation processing methods,a Loopy Belief Propagation(LBP) stereo matching algorithm based on Simple Linear Iterative Clustering(SLIC) image segmentation is proposed.The image is segmented using the LSC algorithm and modeled using a set of plane models so that each segmentation region corresponds to at least one parallax plane label.The self-adapting matching cost calculation parallax is introduced to obtain reliable matching pixels,and the parallax plane estimation is performed by least squares plane fitting,and the LBP algorithm is used to optimize the parallax plane label.Experimental results show that compared with GC+occ,MultiGC and other algorithms,the algorithm has high matching accuracy and can handle low texture areas reas.
-
Computer Engineering. 2018, 44(6): 263-269. https://doi.org/10.19678/j.issn.1000-3428.0049758
Head pose estimation is the key to detect identity and understand behavior in many intelligent systems,but it is influenced by illumination,occlusion and resolution.Aiming at the problem that the accuracy of head pose estimation for color 2-D images is not high enough,based on the analysis of existing face pose estimation methods,a head pose estimation approach based on facial feature point localization is proposed.This approach combines the Adaboost algorithm and the ellipse skin color model to detect the human face,which provides the accurate face area in the picture.Hough circle detection method is used to locate the eyes and nostrils.By using the position information of the eyes and nostrils,the eye and nose positioning results are compared with the eye nose in the face head posture,so that the different head posture is rough estimated.Experimental results show that this approach can achieve a recognition rate of 93.53% on 6 different head poses face outside.
-
Computer Engineering. 2018, 44(6): 270-278. https://doi.org/10.19678/j.issn.1000-3428.0049612At present,many model-driven software safety analysis approaches have been proposed to identify the software failure modes and system hazards of the software products.However,these existing approaches do not integrate the safety constraint into the corresponding models,which can not ensure the analysis effectiveness and sufficiency.To solve the above problems,the safety analysis approach for the embedded software based on the Object Constraint Language(OCL) is proposed in this paper.The constraints related to the safety is described by the OCL representation to obtain the OCL-constraint model for integrating the safety constraint into the UML model.On this basis,considering the characters of embedded software,the failure mode is idenitified.This apporach is applied on the embedded engine control software to illustrate its effectiveness and feasibility.The experimental results show that it can identify the failure modes automatically and improve the effectiveness and sufficiency of the model-driven safety analysis as well as reduce the analysis effort.
-
Computer Engineering. 2018, 44(6): 279-282,287. https://doi.org/10.19678/j.issn.1000-3428.0046961For the open-loop control,strong time delay and low prediction precision in the air input system of bio-oxidation tank,a new method based on the dynamic data driven for predicting air input of bio-oxidation tank is proposed in this paper.Firstly,a state space model is constructed with the time series data of multistage bio-oxidation tanks.Then,Kalman filtering algorithm is used to implement data fusion from the prediction and measured one.And the prediction and model parameters are updated in real time online.Thus,the frame of the prediction model of bio-oxidation tank air input based on the dynamic data driven is constructed.Experimental results show that,the prediction framework can accurately predict the amount of air in the bio-oxidation tank.
-
Computer Engineering. 2018, 44(6): 283-287. https://doi.org/10.19678/j.issn.1000-3428.0047075The traditional Dynamic Framed Slot ALOHA(DFSA) algorithm easily brings the waste of time slot.What is more,the Q value is adjusted slowly.Therefore the throughput of Radio Frequency Identification(RFID) system is lower.Aiming at this problem,based on the analysis of the traditional Q algorithm,this paper proposes an improved anti-collision algorithm which is on the basis of time slot grouping,so as to eliminate above defect.In this algorithm,time slot can be divided into several groups based on the traditional Q algorithm.At the end of each inventory,it dynamically adjusts the Q value according to the estimated number of labels.The simulation results and the experimental results in hardware platform indicate that,the average throughput rate of this algorithm is about seven percent higher than traditional Q algorithm and is about three percent higher than MQ algorithm.
-
Computer Engineering. 2018, 44(6): 288-293,299. https://doi.org/10.19678/j.issn.1000-3428.0048802
At present,commercial wireless chips can only realize the communication function of the industrial wireless network standard,while leaving the functions like the slot scheduling and synchronization to rely on timing interrupt by the software.Therefore,previous solutions increase the difficulty of development and restrict the popularization of industrial wireless protocols such as WIA-PA.To solve the above problems,in this paper,a system level industrial wireless chip for industrial wireless network is designed,named WIASoC2400,which supports the physical layer protocol of the TSCH mode.It bases on the TSCH mode of the IEEE802.15.4e protocol and the ARM Cortex-M3 kernel,and integrates 2.4 GHz WIA-PA wireless communication module.In addition,in order to ensure the security of data transmission,it includes AES-128 encryption and decryption module,which is consistent with the security specification of IEEE802.15.4-2006 protocol CCM* mode.Simulation results demonstrate that WIASoC2400 can satisfy the requirements of time slot synchronization accuracy and frequency hopping in TSCH mode,and has the advantages of precise timing,fast processing and simple implementation.
-
Computer Engineering. 2018, 44(6): 294-299. https://doi.org/10.19678/j.issn.1000-3428.0047392
For the traditional Spatio Temporal Context target tracking(STC) algorithm,no change of tracking’s window for a long time leads to learning space context model does not have a targeted problem when the target scale changes.This paper proposes a Spatio Temporal Context target tracking algorithm for Adaptive Learning(STC-AL).The Scale Invariant Feature Transform(SIFT) is extracted from the front and back output windows and used to eliminate false matches.After analyzing the matching point set,the output window is adjusted,and the learning and updating of the traditional spatial model is improved.Experimental results show that STC-AL algorithm can adapt to changes in the target scale,and tracking is more accurate compared with that of STC algorithm,CT algorithm and KCF algorithm.
-
Computer Engineering. 2018, 44(6): 300-304,311. https://doi.org/10.19678/j.issn.1000-3428.0046410
The main factors that influence the recommendation results in location recommendation system include location,personal interest,social relationship and time cycle.In order to effectively integrate 4 factors to personalized location recommendation,the corresponding selection probability model is constructed for each factor.The influence of 4 factors on user selection is analyzed.Finally,the heuristic recommendation algorithm is proposed by combining 4 factors.Experimental results show that,compared with the traditional location-based recommendation algorithm,the proposed algorithm has better performance and the recommended results are more acceptable to users.
-
Computer Engineering. 2018, 44(6): 305-310. https://doi.org/10.19678/j.issn.1000-3428.0047731
Based on the characteristics of Uyghur language,a method of identifying Uyghur language event coreference relationship based on Stacked Denoising Autoencoder(SDAE) is proposed.This paper divides the Uyghur events to the candidate event pairs,extracted the nine features,basic characteristics of the event,the trigger word and the event distance.At the same time,the word embedding is used to calculate the semantic similarity of Uyghur events trigger words,taking semantic similarity as one of the features.And then training SDAE model,using softmax to complete the identification task of Uyghur language event coreference relationship.Experimental results show that SDAE is more suitable for the identification task than Support Vector Machine(SVM),the shallow machine learning model,and the use of word embedding further enhances the identification performance.
-
Computer Engineering. 2018, 44(6): 311-315. https://doi.org/10.19678/j.issn.1000-3428.0047478
In order to analyze the exhaust emission regularity under the actual traffic conditions of urban roads,the MCD and STCA models of cellular automaton are improved,on this basis,a model based on the speed-acceleration query table coupled with traffic flow cellular automaton is built for vehicle emission statistics.The coupling model is implemented under different traffic flow service levels,and the speed,acceleration and working conditions are obtained.The distribution of the vehicle’s working conditions,three typical exhaust emission processes,and the total emissions of 1 h and 1 km are statistically analyzed.The results show that the coupling mechanism of the model is clear,and it can be applied to analyze the discharge regularity of single vehicle and vehicle flow under the actual traffic service levels.
-
Computer Engineering. 2018, 44(6): 316-320. https://doi.org/10.19678/j.issn.1000-3428.0046745
In multi-target tracking system,some targets will be lost in the tracking when the targets are crossing or closed to each other used by Gaussian Mixture-Probability Hypothesis Density(GM-PHD).In order to solve this problem,an improved algorithm for scenario with crossing targets is put forward.After update,the estimated Gauss label is managed.If the estimated number of targets is reduced,it is needed to determine whether the targets are crossing or closed to each other.If these trajectories are closed to each other,the followed step is label management and weight reset of Gaussian items,and target states and tracks are re-estimated.Otherwise,the reduction of target will be regarded as a normal target extinction phenomenon,and track management is directly carried out.Experimental results show that the improved algorithm can solve the missed detection caused by the crossing targets successfully,and has good stability.
Volume 50 No 3
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