15 January 2019, Volume 45 Issue 1
-
Select all|
-
Computer Engineering. 2019, 45(1): 1-8. https://doi.org/10.19678/j.issn.1000-3428.0052826
With the widespread use of distributed heterogeneous networks in embedded software systems,the complexity of software systems increases geometrically.It is necessary to verify the correctness and sufficiency of system requirements and design schemes in the demonstration stage of system design,and discover system design flaws to avoid possible systemic problems.Therefore,a demand modeling method based on SysML for embedded software system structure and behavior is proposed.Through the hierarchical modeling of the structure and logic behavior of the embedded software system,and using the simulation characteristics of Modelica discrete and continuous fusion,in the system requirements demonstration stage,based on the established simulation model,the key functional indicators of the system are simulated and verified.Results show that the method is feasible and useful for requirements demonstration.
-
Computer Engineering. 2019, 45(1): 9-16,22. https://doi.org/10.19678/j.issn.1000-3428.0049485
Resource Description Framework(RDF) storage scheme based on relational database mostly does not consider the domain characteristics,resulting in insufficient query performance.Therefore,an improved medical record storage scheme is proposed.According to the characteristics of the multi-relationship of the original medical record data,the transformation plan of the RDF triple is designed.The original medical record data has many characteristics such as large null value,large number of predicates and unpredicted predicates.The entity and attributes of the medical record RDF triple are ID-based based on the improved triple storage table storage scheme.On this basis,the entity type table is designed,and the SPARQL-to-SQL query conversion algorithm is used for the medical record.Experimental results show that compared with the type-based storage scheme,the scheme has higher query efficiency.
-
Computer Engineering. 2019, 45(1): 17-22. https://doi.org/10.19678/j.issn.1000-3428.0049905
For the problem that the deep neural network has low accuracy and over-fitting on the mobile platforms,a lightweight Convolutional Neural Network(CNN) architecture is proposed.The 3×3 depthwise separable convolution replaces the standard 3×3 convolution kernel in the SqueezeNet network model basic module Fire,constructs the SparkNet network structure,and replaces the model convolution to obtain the network deformation structure.Experimental results show that compared with the SqueezeNet network structure,the architecture can improve the calculation speed of the network model,effectively reduce the network model size and reduce the number of parameters.
-
Computer Engineering. 2019, 45(1): 23-28. https://doi.org/10.19678/j.issn.1000-3428.0049362
In order to extend the life of equipment for mixed-critical systems,considering instantaneous and permanent errors and taking fault-tolerant solutions to meet safety requirements,a two-stage solution is proposed.The Dynamic Voltage Frequency Scaling(DVFS) technique is used to determine the operating frequency for each task,ensuring that thermal cycling minimizes damage to equipment life.By using the re-execution technology,the equipment life is analyzed under the constraint of reliability constraint and schedulability,and a multi-objective nonlinear programming problem is constructed to obtain the optimal solution.Simulation results show that compared with the four algorithms such as RAND、MFPR、MRPF and DPAS,this algorithm can extend the life of the system by up to 47% under the premise of ensuring system scheduling and reliability.
-
Computer Engineering. 2019, 45(1): 29-34. https://doi.org/10.19678/j.issn.1000-3428.0049296
The prediction of traffic retention is the premise of realizing the automatic timing of intelligent traffic lights.Accurate traffic detention prediction can provide support for the dynamic allocation of traffic signals,thus alleviating urban traffic congestion.A traffic detention prediction system is proposed.The Traffic Flow Prediction (TFP) algorithm based on time series segmentation and Extreme Learning Machine (ELM) is used to predict traffic flow,the simulation scheme of road system is designed,and the predicted traffic is simulated to construct extended secondary Markov randomization model,calculating the predicted value of traffic retention.Experimental results show that compared with BP neural network,the system can accurately predict traffic detention,and can provide theoretical basis for urban traffic guidance and control.
-
Computer Engineering. 2019, 45(1): 35-39. https://doi.org/10.19678/j.issn.1000-3428.0048783
The location of the Wireless Sensor Network(WSN) to obtain the message node needs to be located real time.However,due to the limited energy and poor reliability of the sensor node,a localization algorithm based on Received Signal Strength Indication(RSSI) is proposed,considering the limitation of power consumption and hardware.The location of the unknown node is estimated by the sequence of RSSI values received by the unknown node from the anchor node,and the anchor node moves according to a specific trajectory in the monitoring area,covering all points in the area as much as possible.Simulation results show that the algorithm is feasible and has high localization accuracy.
-
Computer Engineering. 2019, 45(1): 40-44. https://doi.org/10.19678/j.issn.1000-3428.0049245
For the problem of load imbalance of base stations in relay Heterogeneous Networks(HetNets),in order to achieve the best User-Pair Association(UPA) in network,a UPA scheme is investigated for relay-assisted multi-tier HetNets in this paper.For such relay-assisted HetNets,this paper defines the so-called best relay in each tier for a typical mobile user-pair by using the equivalent end-to-end Biased Receive Power(BRP).Then it proposes the Maximum Harmonic Mean-based UPA(MHM-UPA) criterion and models the locations of all network elements as independent Poisson Point Process(PPP).Based on the method of random geometric analysis and by using feasible mathematical analysis,it derives the corresponding UPA probabilities as well as the corresponding statistical description.Analysis results show that this scheme exploits jointly the effect of the both source-relay and relay-destination links,so that the bottleneck effect of dual-hop relay systems is overcome,and it can improve the system performance.
-
Computer Engineering. 2019, 45(1): 45-49,54. https://doi.org/10.19678/j.issn.1000-3428.0048411
Aiming at the problem of unbalanced network energy consumption caused by the unreasonable number and distribution of Cluster Head(CH) in LEACH protocol,a CH selection method based on multi-attribute decision approximation Technique for Order Preference by Similarity to Ideal Solution(TOPSIS) is proposed.By taking into account the residual energy of nodes,number of neighbor nodes,distance to base station and transmission range,a group of CHs is selected by Pareto optimal solution.Then,the optimal CH is selected by TOPSIS.Simulation results show that TOPSIS is superior to LEACH and Analytical Hierarchy Process(AHP) in terms of the number of remaining nodes and network energy consumption,and it can effectively extend the network life cycle and make the network energy consumption more balanced.
-
Computer Engineering. 2019, 45(1): 50-54. https://doi.org/10.19678/j.issn.1000-3428.0049150
Aiming at the non-adaptive problem of selecting the optimal Micro Base Stations(MBSs) and beam according to the channel environment,based on the limited feedback beamforming technique of the random vector quantization codebook,the expression of the maximum average Effective Signal-to-Noise Ratio(ESNR) is derived,and an Adaptive Base Station and Beam Selection(ABSBS) algorithm for user-centric distributed millimeter-wave cellular is proposed.According to the location of the user in the cellular,the optimal MBSs is adaptively selected with the average ESNR as the cost function.The simulation results show that,when the feedback bit number is 6 bit and the Signal-to-Noise Ratio (SNR) is more than 12 dB,the error rate and ergodic capacity of the proposed algorithm are better than those of the fixed number of micro base stations,and the channel gain of 2 dB~4 dB is obtained.
-
Computer Engineering. 2019, 45(1): 55-60. https://doi.org/10.19678/j.issn.1000-3428.0049618
The establishment of Ultra-dense Network(UDN) can increase the system capacity of hotspots,but due to the serious interval interference in the network,the deployment of UDN is limited.Therefore,a power optimization control scheme based on cell clustering is proposed.Clustering is performed by a preset system average interference threshold,and cells with large interference in the system are grouped into one cluster,when the cell in the cluster interferes with other cell users below a set threshold,the cell is removed from the cluster.The power control of the scheme is controlled by the controller according to the weight value occupied by the base station in the cluster,and the corresponding power value is allocated to each base station,thereby achieving the control purpose.Simulation results show that the scheme can effectively suppress the interference in the network and improve the system throughput.
-
Computer Engineering. 2019, 45(1): 61-66. https://doi.org/10.19678/j.issn.1000-3428.0048779
Aiming at the problem of spreading code estimation of multi-rate Long Code Direct Sequence Code Division Multiple Access (LC-DS-CDMA) signal,a matching estimation method based on triple correlation feature information is proposed.Considering the uniqueness of the peak value of the triple correlation function of m-sequence,the triple correlation characteristic information matrix of periodic and aperiodic multi-rate LC-DS-CDMA signals is constructed.According to the length of the received signal,the periodicity of the received signal is judged,and the triple correlation column vector of the corresponding received signal is estimated.By matching with the corresponding characteristic information matrix and normalizing the matching results,the spreading code estimation of each user in the received signal is completed.Simulation results show that the proposed method can effectively estimate the spreading code of multi-rate LC-DS-CDMA signal,and the correct estimation rate is higher than using the blind source separation method.
-
Computer Engineering. 2019, 45(1): 67-72,77. https://doi.org/10.19678/j.issn.1000-3428.0049281
Routing Protocol for Low Power and Lossy Networks (RPL) results in unbalanced energy consumption due to unbalanced load.To solve this problem,a RPL multipath data transmission mechanism is proposed.In the process of network topology construction,the optimal multi-parent node set of each node is selected according to the data transmission cost.Combining with wireless link quality,node residual energy,node cache occupancy rate and the number of relay nodes,a data flow allocation metric is proposed.Based on this metric,a flow allocation strategy is proposed to maximize the load balance of the network to obtain the optimal data transmission scheme.Simulation results show that compared with RPL and ELT-RPL,this mechanism can maximize load balance and node energy consumption balance,prolong network lifetime and improve routing reliability.
-
Computer Engineering. 2019, 45(1): 73-77. https://doi.org/10.19678/j.issn.1000-3428.0049579
In view of the complexity of controller deployment in large Software-Defined Network(SDN),an improved K-means network partition algorithm is proposed with the goal of minimizing control delay.Through the clustering problem modeling controller deployment problem,considering the connectivity of nodes in the real network environment,the shortest path between nodes is used to replace the traditional Euclidean distance to calculate the transmission delay.Experimental results show that compared with the K-means algorithm,this algorithm can effectively reduce the maximum delay and average delay of the network and accurately deploy the controller.
-
Computer Engineering. 2019, 45(1): 78-83,90. https://doi.org/10.19678/j.issn.1000-3428.0048166
In order to solve the problem of low accuracy of node localization algorithm based on Received Signal Strength Indicator(RSSI) in wireless sensor network,a node fuzzy information location algorithm based on Voronoi graph partition is proposed named NFIL-VD.According to the number of anchor nodes,the location area is divided into different Voronoi cell,and the vertex coordinates of each Voronoi cell are obtained.The vertex nodes which can be used as reference nodes are selected by the Gaussian filtering method,and the unknown nodes are located jointly by vertex nodes and anchor nodes.The final location of unknown nodes is calculated by fuzzy information location method.Experimental results show that compared with MANLFI algorithm and FINL-DT algorithm,this algorithm can effectively improve node localization accuracy and reduce network energy consumption.
-
Computer Engineering. 2019, 45(1): 84-90. https://doi.org/10.19678/j.issn.1000-3428.0048977
The introduction of Mobile Data Collector(MDC) in Wireless Sensor Network(WSN) can alleviate energy imbalances and “energy holes” problems in traditional static networks.In order to optimize the energy saving effect of MDC on the network,this paper designs algorithm to allow nodes having a probability of movement in different directions away from or near the sink,and establish data collection trees of different sizes at different network node densities to balance network energy consumption and adjust the location of collectors before data collection to reduce data transmission energy consumption.The MDC motion strategy is given and combined with the DSDV routing protocol,the routing protocol MDSDV with mobile data collector is proposed.Experimental results show that MDSDV can improve the data transmission success rate of the network and prolong the lifetime of the network.
-
Computer Engineering. 2019, 45(1): 91-95. https://doi.org/10.19678/j.issn.1000-3428.0049519
QARMA algorithm is a lightweight and adjustable block cipher algorithm with Substitution Permutation Network(SPN) structure.The ability of QARMA algorithm to resist the related-key impossible differential attack is studied.According to the characteristics of QARMA-64 key arrangement,a 7-round related-key impossible differential differentiator is searched.3 rounds are added before and after the differential differentiator to form the 13-round related-key impossible differential attack.Analysis results show that the proposed attack has more attack rounds,less time complexity and less space complexity than the existing intermediate encounter attack when guessing 52 bit key.
-
Computer Engineering. 2019, 45(1): 96-102. https://doi.org/10.19678/j.issn.1000-3428.0049731
The Location-Based Social Network(LBSN) uses the user’s check-in data to recommend the Point of Interest (POI),but for the consideration of data privacy,various social platforms are unwilling to share data directly.In order to provide a better POI recommendation service by synthesizing the data of various social platforms,a privacy preserving POI recommendation algorithm based on Locality-Sensitive Hashing(LSH) is proposed.The similar user set is selected by LSH,which greatly reduces the computation cost and satisfies the user’s rapid response demand.LSH and Paillier homomorphic encryption techniques are used to protect data privacy from disclosure.Experimental results on real data sets show that the proposed algorithm is superior to the traditional collaborative filtering recommendation algorithm based on users in response time and prediction accuracy.
-
Computer Engineering. 2019, 45(1): 103-108. https://doi.org/10.19678/j.issn.1000-3428.0049298
The Chinese keyword fuzzy ciphertext search scheme relies on a predefined keyword fuzzy set,which has higher search complexity and storage cost.A multi-keyword ciphertext fuzzy search scheme based on pinyin similarity in cloud storage is proposed.The pinyin similarity and Euclidean distance are used to measure the similarity of Chinese characters.Based on the Bloom filter,the local hash function of p-stable distribution is used to construct the index to realize ciphertext fuzzy search of multiple Chinese keywords.Experimental results show that compared with fuzzy search over encrypted cloud data based on keywords,the scheme has high search efficiency,low storage cost and can ensure information security.
-
Computer Engineering. 2019, 45(1): 109-114. https://doi.org/10.19678/j.issn.1000-3428.0049277
The SWIFFTX Hash algorithm with the SWIFFT algorithm as an important component is failed to enter the second round of the SHA-3 campaign due to efficiency issues.In order to explore ways to improve the efficiency of the SWIFFTX Hash algorithm,this paper analyzes the implementation process of the SWIFFT algorithm.By drawing a Fast Fourier Transform(FFT) flow chart,it estimates the addition/subtraction and multiplication of the SWIFFT algorithm.In addition,it proposes a way to calculate the intermediate parameter ω.Analysis results show that:when the storage space is small,the 16-point FFT is more efficient to implement the SWIFFT algorithm;when the storage space is sufficient,the 8-point FFT is more efficient.
-
Computer Engineering. 2019, 45(1): 115-120,128. https://doi.org/10.19678/j.issn.1000-3428.0049576
Aiming at the problem that the user key is easy to be leaked in the existing ciphertext policy attribute encryption scheme,a ciphertext policy scheme based on secure three-party computation protocol is proposed.By using Attribute Authorization center(AA),cloud Data Storage Center(DSC)and User,a secure three-party calculation is carried out to construct the proxy-free key publishing protocol,which enables the client to have the necessary subkeys to generate the complete key.The security analysis shows that this scheme can effectively eliminate the threat caused by the single secret key generation center and the easy leakage of the user key in the process of transmission to the user and enhance the security of the user key.
-
Computer Engineering. 2019, 45(1): 121-128. https://doi.org/10.19678/j.issn.1000-3428.0049571
In order to improve the security and decryption quality of optical image encryption algorithm and facilitate key storage,management and transmission,an image encryption algorithm based on Quick Response(QR) code constraint and chaotic Gyrator transform is proposed.A two-dimensional coupled chaotic system is introduced,and two random phase masks are output after iteration using plaintext pixels.Two chaotic sequences are used to calculate the rotation angles of the two Gyrator transforms,and then the corresponding fast response codes are generated by plaintext.The codes are placed on the input plane of two random phase masks and a real ciphertext is output by two-stage Gyrator transforms.At the same time,the traditional phase retrieval technology is improved by using the support constraints of the QR code,and the original QR code is retrieved from the ciphertext and the decrypted image is decrypted.Experimental results show that the proposed algorithm has higher security and decryption quality than the optical image encryption algorithm based on circular harmonic component expansion and Gyrator transform,under noise attack,the distortion of restored image is lower.
-
Computer Engineering. 2019, 45(1): 129-135,140. https://doi.org/10.19678/j.issn.1000-3428.0049299
The existing Identity-based Encryption with Keyword Search (IBEKS) cannot share search permissions.Therefore,an Identity-based Proxy Re-encryption with Keyword Search (IBPKS) scheme in cloud environment is proposed.The identity-based encryption algorithm is applied to searchable encryption,and the efficient sharing of search authority is realized in the form of agent.At the same time,the formal definition of IBPKS scheme and the security game are given.Experimental results show that the proposed scheme is secure and efficient,and consistent in random oracle model and can resist adaptive identity selection and plaintext attack.
-
Computer Engineering. 2019, 45(1): 136-140. https://doi.org/10.19678/j.issn.1000-3428.0049721
Searchable encryption technology saves user’s local storage resources and simplifies user’s search operation.However,most searchable encryption schemes are only static search.Therefore,this paper presents a dynamic searchable encryption scheme based on bilinear pairings which can delete specified Identity(ID) files.At the same time,the authentication technology is introduced.Before server deletes file,server authenticates user who issued the deletion request and effectively prevents malicious user from deleting maliciously.Analysis results show that the scheme achieves Indistinguishability Under Chosen Plaintext Attack(IND-CPA) security on the basis of the realization of deletion function.
-
Computer Engineering. 2019, 45(1): 141-144,152. https://doi.org/10.19678/j.issn.1000-3428.0049316
In this paper,two improved schemes based on HB protocol are proposed for erasuring attacks on computing resource-limited devices,to make Learning Parity with Noise(LPN)-based cryptographic authentication protocol can resist erasure attacks.The first scheme adds a parity bit to the secret key to detect the erasure attack and then resists it.The second scheme uses Hamming code to encode the secret key.It keeps the original key even the erasure attack works.Experimental results show that both of the two schemes can increase the anti-erasure attack property on the basis of maintaining the security of the original protocol.
-
Computer Engineering. 2019, 45(1): 145-152. https://doi.org/10.19678/j.issn.1000-3428.0048743
The side-channel detection method screens the hardware Trojan circuit by collecting physical parameters such as power consumption,delay,and electromagnetic field,but its detection performance is seriously affected by process fluctuation,and process uncertainty increases as the chip process size shrinks and the subthreshold leakage current increases.Therefore,the influencing factors of hardware Trojan side-channel detection are studied.The ISCAS’85 c880 reference and Trojan circuits are tested in a SMIC 130 nm and 65 nm standard CMOS process by building a side-channel detection model.The results show that optimizing the test vector and appropriately reducing the power supply voltage can reduce the influence of process fluctuation,and the average comprehensive sensitivity is increased by 5.60%(130 nm) and 0.40%(65 nm),respectively.At the same time,using the absolute difference ratio of the static/dynamic test vector group as the auxiliary discriminant basis,the number of test vector screening iterations can be reduced.
-
Computer Engineering. 2019, 45(1): 153-158. https://doi.org/10.19678/j.issn.1000-3428.0049801
The traditional event trigger word extraction methods over rely on Natural Language Processing(NLP) tools in the feature extraction process,which is easy to cause error accumulation.To solve this problem,a new event trigger extraction based on convolutional bidirectional Long Short Term Memory(LSTM)network is proposed.This method utilizes a convolution operation to extract word contextual information and retains sentence-level features through LSTM unit to improve the performance of event trigger word extraction.Experimental results on the ACE2005 English corpus show that the method achieves an F value of 69.5% on the event trigger word recognition and classification stage,which has a good extraction effect.
-
Computer Engineering. 2019, 45(1): 159-164. https://doi.org/10.19678/j.issn.1000-3428.0049128
The partition accuracy of the initial population which is generated by the community mining based on traditional Genetic Algorithm(GA) tends to be low,which can easily lead to the poor overall search performance.To solve this problem,an improved GA is proposed and introduced into community mining study.In the algorithm,the structural similarity and the roulette wheel selection method are adopted so that each gene of chromosomes tends to choose the neighbor node with a larger structure similarity.In this way,the community partitioning quality of the initial population is improved and the convergence rate of this algorithm is accelerated.Experimental results show that in the artificial benchmark network,the partition accuracy and modularity of the initial population generated by the propose algorithm are generally increased by 18% and 12%,compared with the traditional GAs,and the overall partition accuracy of that is averagely increased by 24.02% and 22.01% compared with that of FEC and FN algorithms.In the real world network,the accuracy of community parition of the proposed algorithm is also much better than those of FN,FEC and LPA algorithms.It is verified that the algorithm has better community mining performance.
-
Computer Engineering. 2019, 45(1): 165-171,177. https://doi.org/10.19678/j.issn.1000-3428.0049403
For the fine-grained classification of question texts,which include that the features of text are sparse,the overall features of the text are similar,and the features of local differences are difficult to extract,a classification method based on the combination of semantic expansion and attention network is proposed.The semantic unit is extracted by the dependency syntax analysis tree,and the similar semantic regions around the semantic unit are calculated and expanded in the vector space model.The Long Short Term Memory(LSTM) network model is used to encode the extended text,the attention mechanism is introduced to generate the vector representation of the question text,and the problem text is classified according to the Softmax classifier.Experimental results show that compared with the traditional text classification method based on deep learning network,this method can extract more important classification features and has better classification effect.
-
Computer Engineering. 2019, 45(1): 172-177. https://doi.org/10.19678/j.issn.1000-3428.0048758
Aiming at the multi-predicate phenomenon in semantic role labeling,this paper proposes a Chinese multi-predicate semantic role labeling method from the perspective of Graphical Model(GM).It realizes joint semantic analysis for all predicates in the same sentence and uses Randomized Hill Climbing(RHC) algorithm to optimize GM.The proposed method can fully utilize the interrelations during predicates in the same sentence as the global features to improve the discrimination of semantic roles.Experimental results on Chinese Proposition Bank(CPB) show that the proposed method can significantly improve classification effect of semantic role labeling.
-
Computer Engineering. 2019, 45(1): 178-185,191. https://doi.org/10.19678/j.issn.1000-3428.0049207
Aiming at the problem of advertising Click Through Rate(CTR) prediction on intelligent mobile devices application platform,this paper proposes a novel online advertising business architecture for intelligent mobile devices based on the traditional CTR prediction method on PC Web platform.With this architecture,an online advertising prediction model based on machine learning is designed to integrate and extract the multiple source features such as user information,advertising content and user usage environment,so as to achieve accurate prediction of online advertising CTR.Combined with the characteristics of the mobile APP application environment,the CTR prediction performance is improved by adding the user’s historical behavior data into the prediction model.Experimental results show that this model has a high accuracy rate of CTR prediction.
-
Computer Engineering. 2019, 45(1): 186-191. https://doi.org/10.19678/j.issn.1000-3428.0049148
The facial action coding system defines the concept of facial Action Unit (AU) for facial expression information,but the detection effect of AU strength is poor due to the low degree of discrimination between the levels and the large difference in facial expression between individuals.To this end,mining the strong correlation between AU activation and region,a new feature extraction algorithm based on region and feature fusion is proposed,and an AU strength calculation method is given,high AU strength and low AD strength are classified,the final strength of the AU is judged based on the ordered regression.The algorithm utilizes the strong separability of strong AU and weak AU,considers the correlation between different AU strengtis,and exerts the advantages of classification and regression methods in AU strength detection.Experimental results on the DISFA and FERA2015 datasets show that the proposed algorithm is robust and the AU strengh is better than CNN and VGG16.
-
Computer Engineering. 2019, 45(1): 192-198,205. https://doi.org/10.19678/j.issn.1000-3428.0048673
In order to increase the sparseness of the solution of Least Squares Support Vector Machine (LS-SVM) algorithm and improve its operation efficiency,a variable samples learning LS-SVM algorithm is proposed.Some samples are randomly selected from the training set as the initial working set,and the training process is divided into two stages:sample increment and sample reduction.In the sample increment stage,select specific samples according to KKT conditions to join the working set and train.In the sample reduction stage,Negative Slack Variable Pruning Strategy (NSVPS) and Dual Objective Function Deviation Pruning Strategy(DOFDPS) pruning strategy are used to achieve pruning.On this basis,the residual classifier is used to construct learning classifier.Experimental results show that compared with SMO,SMO-new and ISLS-SVM algorithm,the algorithm has the advantages of high sparsity,fast operation speed and no loss of precision.
-
Computer Engineering. 2019, 45(1): 199-205. https://doi.org/10.19678/j.issn.1000-3428.0049401
Aiming at the problem that traditional musical instrument identification depends on feature selection and elementary acoustical feature,a 5-layer Convolutional Neural Network(CNN) extracting high-level time-frequency information of timbre layer by layer is proposed,whose input is auditory spectrum containing harmonic information and close to human perception.The mono convolution kernel of first layer is improved by multi-scale kernel of time and frequency axises to effectively extract time-frequency information from auditory spectrum.Experimental results on IOWA database show that using the improved multi-scale convolution kernel can achieve 96.95% recognition accuracy,which is better than using a mono convolution kernel.Under the same network structure,the recognition accuracy obtained by using the auditory spectrum is 9.11% and 3.54% higher than the Mel-Frequency Cepstral Coefficient (MFCC) and spectrogram,respectively,and the misclassification rate of percussion instruments and kindred instruments are 2% and 3.1%,which are less than MFCC and spectrogram.
-
Computer Engineering. 2019, 45(1): 206-209,216. https://doi.org/10.19678/j.issn.1000-3428.0049121
Compared with the traditional recommendation model,deep learning can better understand user needs,project characteristics and user-project interaction,so as to more effectively discover the matching relationship between users and projects.The deep neural network is applied to the recommended scene of the internet bar game,and the user’s personal preference is analyzed.According to the change of interest in the time,the game record of the user history is modeled and trained to provide the user with personalized Top-N game recommendation.Based on the deep neural network design training model,the input layer uses the user preference vector processed by the user’s historical behavior data,the hidden layer uses the multi-layer network of the ReLU activation function,and the output layer uses the Logical regression Softmax structure,and finally the L2 specification item is used.The cost function evaluates the reliability of the learned model.Experimental results under the real data set show that with the increase of the hidden layer,the method can significantly reduce the root mean square error and improve recall rate.
-
Computer Engineering. 2019, 45(1): 210-216. https://doi.org/10.19678/j.issn.1000-3428.0049745
Aiming at the lack of semantic connection between double words in Biterm Topic Model(BTM) short text documents,a BTM based on semantic extension of double words is proposed.Considering the semantic relationship between each word in double words,the word vector model is introduced.By training the word vector model,the semantic distance between word in double words is judged,and the BTM is extended according to the semantic distance.Experimental results show that,compared with the existing BTM,this model not only has better short text topic classification effect,but also improves the performance of semantic association and topic meaning clustering between double words.
-
Computer Engineering. 2019, 45(1): 217-220. https://doi.org/10.19678/j.issn.1000-3428.0049076
To solve the disadvantage of VDCN network structure in large scale factors,a new high-precision single image super-resolution method is proposed.The ReLU activation function is replaced with the PReLU activation function,and the number of network layers is increased.The model uses 25 convolution layers with PReLU activation functions to train and test.Experimental results show compared with the VDCN method,this method takes less time and has stable performance.
-
Computer Engineering. 2019, 45(1): 221-225,232. https://doi.org/10.19678/j.issn.1000-3428.0049791
When salient objects exist at image boundaries,most saliency detection models misdetect salient regions as background templates,resulting in poor detection results.To solve this problem,a saliency detection algorithm based on background template optimization is proposed.A selection strategy is designed to remove the salient super-pixel module in the image boundary area,and an improved background template is established to calculate the salient image based on the background.A compact foreground region is obtained from the saliency map,and the saliency map based on foreground is calculated after describing the appearance and location of salient objects.On this basis,two saliency maps are fused and improved by an energy function,and eventually more smooth and accurate saliency maps are obtained.Experimental results show that the proposed algorithm can improve the accuracy and recall rate of target detection compared with SEG,CA and other algorithms.
-
Computer Engineering. 2019, 45(1): 226-232. https://doi.org/10.19678/j.issn.1000-3428.0049513
The selection of tolerance parameters in traditional tolerance mechanism image haze removal algorithm is affected by whether the image contains sky area or not,and the modified transmittance is prone to stratification.To slove this problem,a correction algorithm of tolerance mechanism based on color attenuation prior is proposed.By judging the ratio of the brightness-saturation difference to the original transmittance of each pixel in the image,bright areas such as the sky are extracted,and then the transmittance of the area is magnified and corrected according to the original transmittance value and the brightness-saturation difference to make it approach the true value.Experimental results show that,without changing the tolerance parameters,compared with the image haze removal algorithm based on dark channel prior,the tonal reductivity and Peak Signal to Noise Ratio(PSNR) of the proposed algorithm are improved,and the quality of the restored image is better.
-
Computer Engineering. 2019, 45(1): 233-238,245. https://doi.org/10.19678/j.issn.1000-3428.0049334
An improved Convolution Neural Network(CNN) feature fusion model is proposed to solve the problem of mid-low resolution vehicle type recognition.The feature fusion strategy is adopted to fuse and reuse different low-level features in CNN.Among them,in order to prevent over-fitting,combined with the sparse structure of the network model,the data enhancement method is used to optimize the training data.Analysis and experimental results show that the model can not only produce more distinguishing features,but also avoid the interference caused by environmental factors to a certain extent.Compared with the traditional CNN model,it has higher recognition accuracy.
-
Computer Engineering. 2019, 45(1): 239-245. https://doi.org/10.19678/j.issn.1000-3428.0049109
Building targets in multispectral image have different characteristics at different scales,and the recognition accuracy of traditional Full Convolution Neural network(FCN) is low.Therefore,a building recognition method for multispectral image based on improved FCN is proposed.After expanding the training set by rotating images,the low-level features of images at four rotating angles and at different scales are extracted from the first to the twelfth layers of the network.After normalizing them into images of the same size,the higher-level features are extracted to realize the accurate recognition of multispectral image buildings.Experimental results show that compared with the traditional FCN method,this method can improve the recognition accuracy and recall rate.
-
Computer Engineering. 2019, 45(1): 246-252. https://doi.org/10.19678/j.issn.1000-3428.0049037
Aiming at the problem of spatial information and spectral information fusion in hyperspectral image classification,a joint sparse representation hyperspectral image classification algorithm based on independent space-spectrum residual fusion is proposed.Class-independent spectral angles are used to classify the images,and the initial labels of pixels are obtained,then the neighborhood space of pixels is constructed according to the specific conditions.The global spatial information of the whole image is extracted and introduced into the joint sparse representation model of space-spectrum.The joint sparse representation residual of image spectrum and space is calculated by using a separate spectral information dictionary and a spatial information dictionary respectively.On this basis,the residual fusion algorithm is used to determine image classes.Experimental results show that compared with SVM,KNN and other algorithms,this algorithm can improve the classification accuracy of spectral images and the classification results are more stable.
-
Computer Engineering. 2019, 45(1): 253-258,263. https://doi.org/10.19678/j.issn.1000-3428.0049192
When using traditional three-dimensional(3D) Otsu method to segment forest canopy image,there are some problems,such as poor anti-noise ability and high computational complexity.To slove this problem,an improved forest canopy image segmentation method is proposed based on the 3D Otsu method.The 3D histogram reconstruction is used to denoise and preserve the edge information.The reconstructed 3D histogram is decomposed into three one-dimensional histograms by dimensionality reduction to reduce the computational complexity,running time and storage space.Experimental results show that compared with the traditional 3D Otsu method,this method can effectively segment forest canopy image,and has better robustness,accuracy and real-time performance.
-
Computer Engineering. 2019, 45(1): 259-263. https://doi.org/10.19678/j.issn.1000-3428.0048978
Human action diversity,scene noise,the camera motion angle changes and other factors increase the difficulty of human action recognition.This paper proposes a human action recognition algorithm based on 3D convolution neural network.Firstly,successive 16 frames of the video are divided into a group as the input.Secondly,the input data is multi-channel processed using the gray,gradient-x,gradient-y,optflow-x and optflow-y,which effectively trains the network parameters.Thirdly,the extracted features are obtained using 5-layer 3D convolution and 5-layer 3D pooling to increase time dimension information,Finally,the recognition results are obtained by two full connection layers and the softmax classifier.Experiment is made on the UCF101 database,and the results show that compared with iDT,P-CNN,LRCN algorithms,the proposed algorithm has a higher accuracy of human action recognition and a faster running speed.
-
Computer Engineering. 2019, 45(1): 264-269,277. https://doi.org/10.19678/j.issn.1000-3428.0048705
Using Ricci curvature flow,the original surface panel measure is pushed forward to the target parameter domain to form the initial panel measure.After transforming the initial panel measure or the target panel measure,a series of continuous transformation panel measure sequences are constructed.Then the parameterized sequence clusters of continuous transformation is constructed by calculating the optimal transmission mapping between the panel measures.The transformation experiments are carried out by using Moebius transformation,curvature reinforcement and importance drive,and results show that compared with the Quasi-area method,this method can construct a variety of different parameterized sequence clusters and achieve better special parameterized effect.
-
Computer Engineering. 2019, 45(1): 270-277. https://doi.org/10.19678/j.issn.1000-3428.0049186
Aiming at the background subtraction problem of surveillance video mixed with impact noise during compression sampling,a robust video reconstruction and decomposition model based on Welsch M-estimation and tensor decomposition regularization is proposed.In order to reduce the impact of effect noise on reconstruction performance,Welsch M-estimation is used to replace the mean square error as a cost function to measure the reconstruction error,and a more robust reconstruction model is constructed.Under the tensor framework,the low-rank difference priors of the background in different dimensions and different scenarios are introduced into the background modeling to obtain the reconstruction and decomposition model,and based on half-quadratic theory and multi-block ADMM method,the corresponding optimization algorithm is given.Experimental results show that compared with the algorithms such as SpaRCS and CS-L1PCA,the algorithm can maintain the robustness of video reconstruction and decomposition under the condition of mixed impact noise.
-
Computer Engineering. 2019, 45(1): 278-283. https://doi.org/10.19678/j.issn.1000-3428.0049375
Aiming at the existing problem that the machine learning algorithm used for earthquake prediction is over-fitting and needs a large number of training sets,this paper introduces the danger theory into the application of earthquake prediction.Based on the analysis of a large number of seismic history source data and the combination of of field experts’ experiences,a seismic prediction method using seismology to acquire feature indexes is proposed.Based on the Gutenberg-Ricthter(G-R) law,the distribution of the characteristic earthquake magnitude and the recent earthquake prediction,this method extracts nine seismic characteristic indexes and uses a dynamic danger theory to predict the probability of a major earthquake event in the next month.At the same time,by analyzing the seismic history data of Sichuan province,the danger theory is used to analyze and forecast the seismic characteristic index,and compared with the existing neural network of earthquake prediction method.Experimental results show that the detection probability,accuracy and R score of the proposed method are higher than BP neural network,which indicates that the reliability of it is higher when using the less sample set.
-
Computer Engineering. 2019, 45(1): 284-291. https://doi.org/10.19678/j.issn.1000-3428.0048634
Focusing on the data requirement of intelligent business district’s construction,this paper designs a multi-source data acquisition and comprehensive evaluation system.The system uses the Pyspider framework to collect data.Based on the idea of dichotomy and the rules of polygon,the business district boundary detection and the inner district point determination are conducted and used to set up the mapping rule between Point of Interest(POI) and business district.In this way,the different definition of business district from different data sources can be solved.Furthermore,60 features are extracted from the integrated data to analyze the structure and comparison of business districts.Experimental results show the system can provide data support for various types of analysis models which can provide some decision support for business behaviors.
-
Computer Engineering. 2019, 45(1): 292-296,302. https://doi.org/10.19678/j.issn.1000-3428.0049057
Path searching is the precondition of the ticket income distribution and the passenger flow assignment of urban rail transit network.Only being extended,the urban rail transit network can find path by using path searching algorithm with labeling node,which prolongs the time of paths searching.Considering the influence of the times of in and out of the station to the path-choice,this paper presents the finding K Shortest Paths (KSP) algorithm that applies to urban rail transit network without expanding the network based on the principle of labeling edge.The application of the two algorithms on urban rail transit network of Beijing shows that,compared with the traditional algorithm with labeling node,the proposed algorithm with labeling edge achieves a good effect in computational efficiency,which can save more than half of the time of finding the KSP of the same OD pair.
-
Computer Engineering. 2019, 45(1): 297-302. https://doi.org/10.19678/j.issn.1000-3428.0048741
This paper designs a fast and stable target tracking controller for non-holonomic crawler robots based on visual orientation and ultrasonic ranging.According to the actual parameters and execution ability of the robot,it designs an efficient parameters tuning strategy for the referenced controller,so that the robot can smoothly and gradually reach the ideal posture.On the premise of not increasing the complexity of the controller,the robot’s execution ability can be improved by adjusting the heading angle of the robot with high priority and shortening the distance between the robot and the target with lower priority.Experimental results show that the controller can drive the robot to track the target quickly and smoothly.
-
Computer Engineering. 2019, 45(1): 303-307,314. https://doi.org/10.19678/j.issn.1000-3428.0049583
Based on the pseudo-fractal network,the relationship between the weighting of the power rate distribution and the nature of the network random walk is studied.A weighted method with parameters is proposed.By analyzing the structure and properties of the network,the recursive relationship between the two generations of eigenvalues of the network is constructed,and the eigenvalues and the multiplicity are obtained.On this basis,the random target access time with the preferred random walk on the network is calculated,and the eigenvalue is used to give the weighted spanning tree count of the network.Analysis results show that the proposed method can be applied to the network of deterministic transfer matrices and has good applicability.
-
Computer Engineering. 2019, 45(1): 308-314. https://doi.org/10.19678/j.issn.1000-3428.0049196
The existing news recommendation methods mostly divide user’s interest into interest and not interest,which is difficult to achieve a more detailed description of user’s interest.In order to solve this problem,this paper introduces the idea of emotional space to propose a expressing method of reading interest based on emotion.In details,the Generalized Regression Neural Network (GRNN) is used to analyze eye tracking data and extract the relations between the eye tracking data and reading interest.And then,a user interest model is proposed based on the analyzing results.Experimental results show that the prediction accuracy of the proposed model is 86%,which is 7% higher than that of the PLSR model.It means that the proposed model has a good practicality.
-
Computer Engineering. 2019, 45(1): 315-320. https://doi.org/10.19678/j.issn.1000-3428.0048022
Aiming at the problem of repeated content transmission and long access delay in the Vehicular Ad Hoc Network(VANET),a hybrid collaborative caching strategy based on the similarity of Pending Interest Table(PIT) is designed.According to the PIT similarity of the central node,a method for adaptively building a caching community is given.The Virtual Distributed Hash Table(VDHT) is used to implement the buffer scheduling of similar areas,and the Leave Copy Down(LCD) strategy is used outside the area to achieve hybrid clooaboration.Experimental results show that compared with the classical ICN cache algorithm,this strategy can reduce network load and node caching operation frequency with a small amount of communication and storage cost,and improve caching utilization.
Volume 50 No 4
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