15 April 2019, Volume 45 Issue 4
-
Select all|
-
Computer Engineering. 2019, 45(4): 1-5. https://doi.org/10.19678/j.issn.1000-3428.0050383
The Jiangmen Underground Neutrino Observatory(JUNO) is a multipurpose neutrino experiment designed to determine neutrino mass hierarchy and precisely measure oscillation parameters,whose experimental data is large and complicated in processing.It is necessary to analyze and process the data by means of a unified and efficient offline computing platform.So,this paper builds the customized Docker images with different operating systems for JUNO jobs by packing all software packages and dependency libraries which JUNO computing needs,and submits various types of JUNO jobs to scheduler to achieve resource sharing.A comparison test of CPU and I/O performances and JUNO job running effectiveness among bare metal,virtual machine and container test environments is done.Results show that the Docker container is capable of JUNO offline data processing and has a lower performance penalty than virtual machine.
-
Computer Engineering. 2019, 45(4): 6-12. https://doi.org/10.19678/j.issn.1000-3428.0050084
To address the requirements of the embedded systems for functionality and real-time performance,this paper presents a dual operating system architecture.Based on the trusted execution environment provided by ARM TrustZone,the dual operating system architecture is built to run the General Purpose Operating System(GPOS) and the Real-time Operating System(RTOS) on the same platform.The tasks in the embedded system are divided into real-time tasks and non-real-time tasks.Real-time tasks run on the RTOS and non-real-time tasks run on the GPOS.Finally,a prototype system is implemented on a physical development board and the performance of the system is evaluated using real-time test tools.Experimental results show that the proposed architecture guarantees not only the functionality of GPOS but also the real-time performance of the system.
-
Computer Engineering. 2019, 45(4): 13-17,24. https://doi.org/10.19678/j.issn.1000-3428.0050008
In the Software Defined Network(SDN) architecture,virtual network mapping is the key technology to achieve network virtualization.To solve the problems of high cost of mapping and long running time,a virtual network mapping algorithm named Simplex-VNM is proposed.In the stage of node mapping,virtual nodes are ranked according to resources requirements and search order,then the mapping node is chosen according to the node connectivity and mapping costs.In the stage of link mapping,the process is equivalent to the minimum cost flow problem which can be solved by the network simplex algorithm.Experimental results show that,compared with NA-PVNM and Improved-vnmFlib algorithms,the proposed algorithm has lower mapping cost and shorter running time.
-
Computer Engineering. 2019, 45(4): 18-24. https://doi.org/10.19678/j.issn.1000-3428.0049794
Aiming at the problem that the traditional offline business traffic identification method consumes a long time,and has poor real-time performance,through the management and control of spatial information network and the efficient arrangement of network resources,a business identification technology of spatial information network based on Software Defined Network (SDN) architecture is proposed.The OpenFlow protocol is used to collect business traffic online,and extract the first five data packets in the flow as a sub-flow,and implement online business classification based on machine learning on the SDN controller.At the same time,a collaborative training algorithm Dif-TriTraining with noise filtering function is presented.Experimental results show that compared with the traditional Tri-Training algorithm,the algorithm can effectively improve the accuracy of business identification.
-
Computer Engineering. 2019, 45(4): 25-29. https://doi.org/10.19678/j.issn.1000-3428.0050025
Matrix operations on vector processors face the problems of low utilization of hardware resources and low data processing capacity.Therefore,based on HX Digital Signal Processor(HXDSP) platform,combined with Discrete Cosine Transform(DCT) algorithm in High Efficiency Video Coding(HEVC) standard,the matrix multiplication is realized by data compression vector method,in order to make full use of the hardware resources of vector processor.Experimental results show that this method can reach the peak value of 32 GMACs of fixed-point multiplication and accumulation operation ability of HXDSP,and the data processing ability can reach 2 GPixel/s,which meets the performance requirements of HEVC coding standard.
-
Computer Engineering. 2019, 45(4): 30-35. https://doi.org/10.19678/j.issn.1000-3428.0049764
In order to deploy multi-controllers reasonably in Software Defined Networking(SDN),considering the three aspects of time delay,load and cost,the average flow installation time delay of switches and the controllers load variance are modeled.A multi-controllers deployment model and its solving algorithm are proposed,and the optimal controllers deployment scheme is solved based on Particle Swarm Optimization(PSO) and simulated annealing algorithm.Simulation results show that when the same number of controllers are deployed,the average flow installation time delay and the controllers load variance of the proposed algorithm are smaller than those of the PSO and the random deployment algorithm.
-
Computer Engineering. 2019, 45(4): 36-40. https://doi.org/10.19678/j.issn.1000-3428.0051365
In order to improve the link bandwidth utilization and throughput of Data Center Network(DCN),a Software Defined Network(SDN) based Flow Probability Path Selection(FPPS) method is proposed.On the basis of the shortest path selection method,the SDN controller is used to classify the network flow,calculate the bandwidth occupation condition of each data flow in the scheduling process.Then,it can reduce the data link fragmentation,optimize the DCN data flow scheduling,and realize the network load balancing.Experimental results show that this method can reduce network transmission delay,improve link utilization and load balance degree.
-
Computer Engineering. 2019, 45(4): 41-48,55. https://doi.org/10.19678/j.issn.1000-3428.0052224
In view of the current situation that avionics networks can not meet the needs of aviation formation cooperative operations in complex battlefield environments,the integrated modular system avionics network and software-defined network are introduced.From the perspective of the practical application of aviation formation cooperative operations,it analyzes the limitations of traditional avionics network architecture under formation cooperative operations.Based on the information interaction requirements of aviation formation cooperative operations,a software-defined multi-platform avionics network architecture is constructed.Based on this,the data plane and control plane structure of the architecture are designed and described.Simulation results show that the network architecture can make the information receiving platform get higher value in unit time,and ensure that the information quickly converges to the current execution task,and provide theoretical support for software defines multi-platform avionics network for aviation formation cooperative operations in future battlefields.
-
Computer Engineering. 2019, 45(4): 49-55. https://doi.org/10.19678/j.issn.1000-3428.0049941
In order to meet the urgent need of improving the management level of the Weihe river basin under the current situation,on the basis of the new station of Beidou and the integration of existing information resources in the Weihe river basin,constructing the integrated ground-air-space monitoring system,is constructed real-time data acquisition of Beidou satellite,remote sensing and ground monitoring is realized by the establishment of water resources monitoring data center of Weihe river basin,water resources monitoring system is designed and developed in Weihe river basin based on development of 3D visualization component by using Tile Pyramid,massive data cache and 3S integration technology.The result of a case study of Shaanxi section of Weihe river basin shows that the water resources monitoring system can provide information service,calculation service and decision service for the management of Weihe river basin based on the integration of massive,multi-source and heterogeneous data.The constructed 3D GIS platform has the characteristics of good visualization effect,simple operation and good scalability,which can provide auxiliary decision support means for the river basin management department.
-
Computer Engineering. 2019, 45(4): 56-60. https://doi.org/10.19678/j.issn.1000-3428.0049374
Millimeter wave communication has a large bandwidth and can achieve Gigabit transmission rate,which will become the main way of 5G mobile communication backhaul.Millimeter wave wireless backhaul in Ultra-Dense Networks(UDN) is studied to optimize routing and link scheduling in backhaul networks with the goal of maximizing network throughput and the number of links satisfying QoS requirements.The optimization problem is modeled mathematically and the transmission path is obtained by linear programming.On this basis,a scheduling algorithm based on Maximum QoS Transmission Set(MQTS) is proposed.Simulation results show that compared with the TDMA and STDMA algorithm,the throughput and the number of successful scheduling links of the proposed algorithm are higher.
-
Computer Engineering. 2019, 45(4): 61-65. https://doi.org/10.19678/j.issn.1000-3428.0050404
Based on the Maximum-Maximum User Pair Association(MM-UPA) criterion,the outage performance of two-hop relay cooperative heterogeneous networks is studied.By means of stochastic geometry and probability statistics,the spatial location of network nodes is modeled as a Homogeneous Poisson Point Process(HPPP),and the analytical expression of outage probability in two-hop relay cooperative heterogeneous networks is derived.On this basis,the impact of outage threshold and bias factor on system outage probability is simulated and analyzed.Results show that when the system deploys MM-UPA scheme,increasing the bias factor of tier k(k≥2) can effectively reduce the outage probability of tier 1 and improve system stability.In addition,the outage probability of single-hop network is smaller and the system stability is higher than that of two-hop networks.
-
Computer Engineering. 2019, 45(4): 66-71,77. https://doi.org/10.19678/j.issn.1000-3428.0048728
Indoor WiFi location accuracy is low,and Pedestrian Dead Reckoning(PDR) location has the problem of cumulative error.Therefore,a WiFi-PDR fusion location algorithm based on Extended Kalman Filter(EKF) is proposed.WiFi achieves matching location by improved WKNN algorithm,and the relative deviation of the Received Signal Strength Indication(RSSI) between the location point and the K-nearest neighbor point is used for weight correction.PDR location adopts gait detection with multiple constraints and online step size estimation.On this basis,EKF is used as the fusion filter of WiFi and PDR location to reduce the WiFi location bounce and the cumulative error of PDR to improve the location accuracy.Experimental results show that the location accuracy of the fusion algorithm can reach 1.8 m under the condition of multiple traveling turns.
-
Computer Engineering. 2019, 45(4): 72-77. https://doi.org/10.19678/j.issn.1000-3428.0049763
A 32 bit parallel concatenated multi-mode polar codes encoder structure is proposed to obtain a polar codes encoder which support multiple codes lengths and have higher throughput.32 rows of the generated matrix are updated in each clock cycle,and 32 bit parallel coding is performed to speed up the coding.The structure of the encoder is simplified by a cascade of two-level coding structures to support 64 bit to 4 096 bit polarization coding.Experimental results show that on Xilinx XC6VLX240t FPGA,the main frequency of the proposed encoder is 303.82 MHz and the throughput is 9.72 Gb/s,compared with Fast Fourier Transform(FFT) structure,its register and lookup table resources are 77.6% and 63.3% lower,respectively,the main frequency can reach 0.796 GHz and the throughput can reach 24.615 Gb/s under 65 nm CMOS process.
-
Computer Engineering. 2019, 45(4): 78-81. https://doi.org/10.19678/j.issn.1000-3428.0050150
The exisiting polar code construction scheme is only applicable to Binary Erasure Channel (BEC) and Additive White Gaussian Noise(AWGN) channel.Therefore,the constrution scheme of polar code in Binary Symmetric Channel (BSC) is studied.The construction method of the polar code and the principle of the coding and decoding algorithm in the BSC are analyzed.The error correction efficiency of the polar code constructed in the BSC is tested by the Successive Cancellation(SC) decoding algorithm.Simulation results show that the polar code constructed by Monte Carlo method has better error correction efficiency than the polar code constructed by Bhattacharyya method.
-
Computer Engineering. 2019, 45(4): 82-86,92. https://doi.org/10.19678/j.issn.1000-3428.0050342
TCP-Illinois is a congestion control algorithm for wired high-speed network.When applied to wireless networks,there is a serious problem of random packet loss caused by noise interference.Therefore,an improved congestion control algorithm TCP-NewIllinois is proposed.The packet loss is determined by the backlog of bottleneck queues.When congestion occurs,the congestion window and slow start threshold are adjusted by bandwidth estimation to improve the bandwidth utilization of wireless network.Simulation results on NS3 platform show that compared with TCP-Illinois algorithm,TCP-NewIllinois algorithm has higher wireless network throughput and transmission efficiency.
-
Computer Engineering. 2019, 45(4): 87-92. https://doi.org/10.19678/j.issn.1000-3428.0049332
In order to ensure the stability of cognitive radio network routing,reduce the number of channel switching and decrease end-to-end delay,a routing protocol based on joint channel selection is proposed.The protocol takes channel availability and switching delay as channel selection criteria.On the premise of ensuring channel availability,it chooses the shortest end-to-end delay route for data transmission to improve the delay performance and data delivery rate of routing.Simulation results show that the proposed protocol has a better performance in terms of packet switching delay and data delivery rate compared with TDRP and PUC-JRCA protocol.
-
Computer Engineering. 2019, 45(4): 93-99. https://doi.org/10.19678/j.issn.1000-3428.0050261
Long Term Evolution(LTE) performs one-way compatible to Worldwide Interoperability for Microwave Access(WiMAX),but the forwards compatibility from WiMAX to LTE cannot be realized.To solve the problem,an efficient uplink scheduling algorithm based on Quality of Service(QoS) is proposed.This algorithm provides elaborate mapping rules for both flows of service,and present dynamic schemes for uplink resource assignment based on priorities.It not only meets the requirements for different links,also ensure fairness among intra-class links.Simulation result shows that the algorithm has an obvious advantage in throughput and QoS comparing to a typical Round Robin(RR) algorithm,shows robustness in overload situation.The interoperability from WiMAX to LTE is successfully achieved.
-
Computer Engineering. 2019, 45(4): 100-107. https://doi.org/10.19678/j.issn.1000-3428.0049825
With the development of the civilian Unmanned Aerial Vehicle (UAV) technology,the security problem for UAV is gradually exposed.Based on the principle of control link spoofing attack,it is proved that the security intensity of remote control data is too low for civilian UAV by an attack example,and a remote control data protection method based on encryption algorithm is proposed.The selection principle of encryption algorithm is discussed,and the feasibility of three encryption algorithms AES,PRESENT and RC4 are analyzed from two aspects of security and performance.The AES algorithm is selected to encrypt the remote control data packet in the control link of UAV system.Finally,the effectiveness of the method is analyzed quantitatively by using the coefficient of difference and correlation between different remote control data.Experimental results show that this method can effectively improve the safety of remote control data for civilian UAV.
-
Computer Engineering. 2019, 45(4): 108-113,118. https://doi.org/10.19678/j.issn.1000-3428.0050452
Aiming at the problem that the traffic anomaly detection method has low detection accuracy and poor adaptability to dynamic changes of network environment,an adaptive threshold network traffic anomaly detection method is proposed according to the strong correlation characteristics of network traffic in adjacent time periods.The sliding window is used to control the number of KL distance values,and the Exponentially Weighted Moving Average(EWMA) model is used to obtain the predicted value of the KL distance at the next moment,and the KL distance subsequence and the predicted value divided by the sliding window are determined.It adapts the threshold range and determines network traffic anomaly detection by determining whether the observed value is within the adaptive threshold range.Experimental results show that the method can effectively detect network traffic anomalies and has high detection accuracy.
-
Computer Engineering. 2019, 45(4): 114-118. https://doi.org/10.19678/j.issn.1000-3428.0049695
In order to solve the edge weight privacy leakage problem of social network,a privacy protection algorithm for weighted social network is proposed.The undirected weighted graph is used to represent the social network,the edge weight sequence is treated as an unassigned histogram,and the weight containing sensitive information is added to the Laplace noise to satisfy the differential privacy protection requirement.In order to reduce noise,the buckets with the same count in histogram are merged into groups,and the differential privacy protection requirements are guaranteed according to the k-indiscernibility between groups,and the shortest path of the network is kept unchanged by consistent reasoning on the original weight sequence.Theoretical analysis and experimental results show that the proposed algorithm can meet the requirements of differential privacy protection and improve the accuracy and practicability of information release.
-
Computer Engineering. 2019, 45(4): 119-123,129. https://doi.org/10.19678/j.issn.1000-3428.0049685
The existing detection method of malicious control command of distribution network is based on the operating rules of power system,but the rules are difficult to maintain and the rule matching takes a long time.According to the context consistency relationship between the uplink measurement information of the distribution network and the downlink control instructions,a malicious control detection algorithm based on word2vec is proposed.Simulating various working conditions on the power distribution network simulation platform and obtaining the labeled data set,the results show that the algorithm can achieve 100% accuracy and 87.2% recall rate,and has high detection accuracy.
-
Computer Engineering. 2019, 45(4): 124-129. https://doi.org/10.19678/j.issn.1000-3428.0050368
In order to improve the availability of published data under the same privacy budget,an improved privacy data publishing method PrivBayes_Hierarchical is proposed based on PrivBayes.Based on the idea of Bayesian network privacy data publishing method,semantic tree is introduced to abstract the data attributes with hierarchical relationships,and Bayesian network are used to describe the dependencies between data attributes.Using Graycode to reduce the impact of random noise on the data accuracy and to optimize the Bayesian network structure learning method to reduce unnecessary privacy budget consumption and improve the availability of data.Experimental results show that this method can obtain higher data precision than PrivBayes for public data and improve the availability of private data sets.
-
Computer Engineering. 2019, 45(4): 130-135. https://doi.org/10.19678/j.issn.1000-3428.0050224
Aiming at the limitation that most traditional public key encryption mode can only send messages from a single sender to a single receiver,a one to one homomorphic encryption algorithm based on integer polynomial ring is designed based on integer homomorphic encryption scheme.On this basis,by modifying the key generation method of one to one homomorphic encryption algorithm and expanding the number of encryption parties,a fullly homomorphic encryption algorithm based on integer polynomial ring for multi-party encryption and one-party decryption is proposed.The correctness and homomorphism of the algorithm are proved,and the security of the algorithm is proved based on Sparse Subset Sum Problem(SSSP) and Approximate Greatest Common Divisor(AGCD) problem under the random oracle model.Performance comparison results show that this algorithm extends the number of encryption parties and improves the efficiency of decryption.
-
Computer Engineering. 2019, 45(4): 136-141. https://doi.org/10.19678/j.issn.1000-3428.0052360
Most of tag ownership transfer protocols are only applicable to the single tag ownership transfer,and are not suitable for the tag group transfer at one time.To solve this problem,a Radio Frequency Identification(RFID) tag group ownership transfer protocol based on trusted third party is proposed.Based on ensuring the bidirectional authentication between the new and old owners and the tags,the communication messages are encrypted by the cross-bit operation and the word synthesis operation.In the process of encrypting the communication messages,the random numbers are mixed,so that the front-to-back communication messages are different,and the crack is increasingly difficult.Based on the participation of trusted third party,without the original owner,the new owner can update the private key directly with the tag to ensure the backward security of the protocol.Performance analysis results show that this protocol has higher security and lower cost.
-
Computer Engineering. 2019, 45(4): 142-147. https://doi.org/10.19678/j.issn.1000-3428.0052276
Aiming the problem that the traditional decision tree classification algorithm in intrusion detection system can only deal with discrete data,an improved intrusion detection method is proposed.The cloud model is used to discretize the continuous attribute of datasets and the genetic algorithm is used to introduce the weighted selection probability function so that the decision tree classification algorithm can detect the attack of DoS,R2L,U2R and PRB.Experimental result of the KDDCUP 99 dataset shows that this method has better intrusion detection and classification performance compared with detection method based on Bayes,Support Vector Machine(SVM) and cloud model discretization.
-
Computer Engineering. 2019, 45(4): 148-156. https://doi.org/10.19678/j.issn.1000-3428.0049992
Aiming at the problem that the Deterministic Finite Automation(DFA)-based matching engine will explode under large-scale and complex rules,a regular expression substring extraction algorithm is proposed.A substring extraction based regular matching engine is designed by applying the substring extraction algorithm to the DFA state explosion scenario.Experimental results show that the algorithm can run on a single rule up to 10 ms,and the decimation rate is as high as 99%.At the same time,the matching engine has better stability and scalability,and the matching speed is better than the related open source matching engine.
-
Computer Engineering. 2019, 45(4): 157-162,168. https://doi.org/10.19678/j.issn.1000-3428.0050398
Traditional machine learning method always needs high cost manual marking process,and exhibits weak ability of generalization in classification task.In order to solve these problems,a label combination semi-supervised learning algorithm is proposed.Taking advantage of the principle of ensemble learning,the algorithm uses the labeled data to train multiple weak learners,and combine them to enhance the generalization ability.Predict the unlabeled data to generate noise labels,and then combine and model these noise labels to make the model more robust.Under the framework of risk minimization,the model converges to the optimal state.Experimental results show that,compared with some existing learning algorithms like Support Vector Machine(SVM),Classification and Regression Tree(CART),Neural Network(NN),the algorithm has relatively good generalization ability.
-
Computer Engineering. 2019, 45(4): 163-168. https://doi.org/10.19678/j.issn.1000-3428.0049468
Aiming at the problem of mismatched point pairs in the point cloud registration process,a double threshold elimination algorithm based on point cloud geometry feature is proposed.According to the geometric characteristics of the point cloud,the translational rotation invariance is obtained during the rigid body transformation process.On the basis of the initial matching point set,the nearest neighbor points of each query point are selected by the k-nearest neighbor method to form a three-point pair.According to the characteristics of the distance between points,the initial pairing is completed.On this basis,the geometrical features of the local region of the three-point pair are described by surface variation,and the final screening of the matching points is completed by analyzing the covariance matrix of the three-point pair.Experimental results show that the method can effectively eliminate the wrong matching points and has higher registration accuracy.
-
Computer Engineering. 2019, 45(4): 169-174,180. https://doi.org/10.19678/j.issn.1000-3428.0050338
In the task of Micro-Blog sentiment analysis,the deep neural-based models are difficult to make full use of the sentiment information.To solve this problem,a Multiple Features Convolutional Neural Networks(MF-CNN) model is proposed.The emotional information in sentences is effectively utilized by combining the abstract features of words and two kinds of calculation methods of neural model input matrix,and then the sentiment classification result is optimized.The sentiment analysis is carried out on COAE2014 and Micro-Blog text data set,and the results show that the classification effect of MF-CNN model is better than that of traditional classifier and deep Convolutional Neural Network(CNN) model.
-
Computer Engineering. 2019, 45(4): 175-180. https://doi.org/10.19678/j.issn.1000-3428.0051647
Aiming at the problem that the existing social friend recommendation algorithm only focuses on people and ignores things,a recommendation algorithm based on association rules and similarity is proposed.By calculating the similarity of the information released by the user every day,the information with the similarity reaching the given threshold is regarded as a transaction record,the published information is regarded as the transaction item,the information database is regarded as the transaction database,and the second-order candidate set is calculated.It is recommended that the publisher who supports the top N items of the highest number of friends be friends.Experimental results show that compared with the friend recommendation algorithm based on association rules and labels,this algorithm has higher accuracy.
-
Computer Engineering. 2019, 45(4): 181-188. https://doi.org/10.19678/j.issn.1000-3428.0049737
Existing entity augmentation techniques can only return a single result,and they are only applicable for augmenting entities with a single attribute.As for entities having multiple attributes,they will return inconsistent results.Two kinds of top-k entity augmentation algorithm are proposed to settle the problems.According to the consistent matching degree,k answer table sets with the highest consistency supporting degree were found in many Web tables to settle entity inconsistent problem.Experimental results show that these two algorithms have implemented top-k entity augmentation with high result consistency and accuracy.The diversity of the algorithm based on consistent matching degree is higher,while the algorithm based on branch-and-bound is more effective in getting reliable entity augmentation results.
-
Computer Engineering. 2019, 45(4): 189-195. https://doi.org/10.19678/j.issn.1000-3428.0049421
Aiming at the problem that it is difficult to balance the global optimization ability and local search ability for a single population algorithm,an inter-species dual-system cooperative bat optimization algorithm is proposed.The whole population is divided into detection system and development system,they evolves and coordinates through information exchange.In the algorithm,the dynamic change operator is designed and applied to realize the real-time balance between global optimization and local optimization.The position update operator is gived to reduce the impact of the randomness of the development system to improve the development efficiency of the local area.The pseudo mutation operator is designed to child maintain the diversity of the detection system to improve the efficiency of the global search.Experimental results show that the algorithm can avoid the local optimization while fast convergence,and can be used to solve the complex optimization problem of multi-local extremum.
-
Computer Engineering. 2019, 45(4): 196-204. https://doi.org/10.19678/j.issn.1000-3428.0050486
Hilbert-Huang Transform(HHT) is an effective method to deal with Electroencephalography(EEG) that includes two parts:Empirical Mode Decomposition(EMD) and Hilbert transform.However,the EMD cannot decompose a signal of low energy and will produce bad Intrinsic Mode Functions(IMF) in the low frequency region.To eliminate the effects of EMD,this paper proposes an EEG processing method which combines Wavelet Packet Transform(WPT) and HHT.Firstly,the EEG is decomposed into a set of narrow-band signals by WPT,the Hilbert energy spectrum of EEG is obtained by HHT,and the average value of the instantaneous energies is calculated as the EEG feature and packaged into a feature matrix.The feature matrix is trained and classified by mixed emotion recognition model which consists of Convolutional Neural Network(CNN),Recurrent Neural Network(RNN),and Support Vector Machine(SVM).Experimental results show that the average recognition rate and the best recognition rate of the four emotions which are happiness,sadness,calmness,and fear are 86.22% and 93.45%.
-
Computer Engineering. 2019, 45(4): 205-210,216. https://doi.org/10.19678/j.issn.1000-3428.0050169
Traditional Topics over Time(ToT) models usually ignore label meta-information in the original data.To solve this problem,an improved ToT model based on label is established.Aiming at the problem that traditional weighting algorithms ignore the influence of vocabulary distribution among and within document sets on weights,combined with the characteristics of document titles,an improved TF-IGM algorithm is used to analyze the weights to extend the generation process of the model.Based on the ToT model,the label attributes of the original document are introduced to construct the improved model and estimate its parameters using Gibbs sampling algorithm.Experimental results show that the proposed model has higher generalization ability than the ToT model.
-
Computer Engineering. 2019, 45(4): 211-216. https://doi.org/10.19678/j.issn.1000-3428.0050129
In order to solve the problem that the probability of zero elements in random pooling is zero,it cannot be selected.An improved hybrid probability stochastic method is proposed.The elements in the pooled domain are deduplicated and sorted in ascending order,and then the power of the corresponding order is added to obtain the weight probability of the element.On this basis,the pooling value is given based on the multi-distribution sampling.Experimental results on the datasets MNIST,CIFAR-10,and CIFAR-100 show that the method in the classification accuracy of the three datasets is 99.50%,72.25%,and 39.05%,respectively.Compared with the traditional pooling method,the method has good classification effect and robustness.
-
Computer Engineering. 2019, 45(4): 217-222. https://doi.org/10.19678/j.issn.1000-3428.0049933
In order to improve the timeliness of 3-D maximum entropy method for image segmentation,an image segmentation method based on equivalent 3-D entropy and Whale Optimization Algorithm(WOA) is proposed.An equivalent 3-D entropy method without logarithmic operation is designed to solve the optimal threshold based on WOA algorithm to avoid exhaustive search and improve the efficiency of the algorithm.Experimental results show that compared with the traditional 3-D maximum entropy method and equivalent 3-D maximum entropy method,this method can reduce the running time without affecting the segmentation effect,and has obvious advantages in noise resistance and image details preservation.
-
Computer Engineering. 2019, 45(4): 223-227. https://doi.org/10.19678/j.issn.1000-3428.0050445
The existing bitewing radiography image segmentation method based on U-Net divides the X-ray image of the bitewing radiography into caries,enamel,dentin,pulp,crown,prosthesis and root canal,but the segmentation accuracy is low.So,this paper proposes an improved method to segmentation bitewing radiograpy images.The conditional Generative Adversarial Network(cGAN) combined with U-Net to segmentation the bitewing radiograpy images.It optimizes the discriminator and the generator to obtain a segmentation feature map with more context information.Experimental results show that the Dice coefficient of the improved method is improved by 0.133 compared to the U-Net method, and the segmentation accuracy is higher.
-
Computer Engineering. 2019, 45(4): 228-234. https://doi.org/10.19678/j.issn.1000-3428.0050656
Because the depth image acquired by the mainstream sensor has image quality problems such as missing depth information area and noise,a depth image enhancement algorithm based on Static/Dynamic(SD) global optimization model is proposed.The SD global optimization model smoothing term is modeled by non-convex functions,which makes it more robust to outliers.The Alternating Direction Multiplier Method(ADMM) based on Hierarchical Identification(HI) is used to solve the SD global optimization model.The method decomposes the objective function into multiple sub-objective functions,and solves each sub-objective function one by one through the HI idea to reduce the complexity of the solution.Experimental results show that the proposed algorithm can effectively remove image noise and suppress depth artifacts while speeding up convergence.
-
Computer Engineering. 2019, 45(4): 235-240. https://doi.org/10.19678/j.issn.1000-3428.0049800
Because the low-cost Unmanned Aerial Vehicle(UAV) does not have a sophisticated inertial navigation system,the shooting area may be some unmanned areas where it is difficult to lay out control points,the conventional aerial photogrammetric processing method cannot be used to obtain the splicing image of the shooting area.Therefore,a method of full-automatic splicing images based on 3D reconstruction is proposed.Dense point cloud of the shooting area is generated by using Structure From Motion(SFM) and Multi-view Stereo(MVS) algorithms,a point coordinate interpolation algorithm based on adjacent point distribution constraint is used to interpolate the space coordinates to be undetermined,and images are geometric-rectified by using indirect differential correction algorithm,so a geometric-consistent splicing image is obtained.Experimental result shows that the method requires no human intervention and the process takes a short time,and the geometric splicing accuracy between adjacent images is about 2.5 pixels.
-
Computer Engineering. 2019, 45(4): 241-247. https://doi.org/10.19678/j.issn.1000-3428.0049463
The traditional total variation algorithm is mostly affected by the staircase effect in the variation process,so it causes texture loss and over-smoothing in the reconstructed image.Therefore,a reconstruction algorithm based on improved non-local means is proposed.The image texture information is preserved by introducing a fractional step model,and the Lagrangian gradient operator is updated by the non-local means filtering method,thereby reducing the computational complexity.Experimental results show that compared with the traditional TVAL3 algorithm,this algorithm can effectively reduce the running time and has better reconstruction performance.
-
Computer Engineering. 2019, 45(4): 248-253. https://doi.org/10.19678/j.issn.1000-3428.0052350
In view of the poor performance of the existing saliency detection method in object highlighting and background noise suppression,a salient object detection algorithm is proposed.An improved Fully Convolutional Neural Network(FCNN) is constructed to capture the semantic information in the image,which can generate the high-level semantic preliminary saliency maps.The semantic knowledge is used to optimize the Manifold Ranking (MR),and then achieve the edge propagation of salient object.The saliency maps under different scales are fused to realize salient object detection.Experiments are carried out on ECSSD,DUT-OMRON and SED2 datasets,the results show that compared with algorithms such as Maximum Symmetric Surround(MSS)and Principal Component Analysis(PCA),the salient objects detected by the algorithm in this paper are more complete and the detection results are more robust in complex scenes.
-
Computer Engineering. 2019, 45(4): 254-261. https://doi.org/10.19678/j.issn.1000-3428.0051769
Aiming at the defect problem in the low-level characteristics of pulmonary nodules in the process of network transmission,an improved U-Net convolution neural network algorithm based on multi-scale feature structure for pulmonary nodule detection is proposed.High-level features are acquired by convolution and pooling operations,high-speed flow of feature information between input and output layers is achieved by using dense network,and multi-scale features are generated combining with expanded convolution to improve the utilization of low-level features of pulmonary nodules.Experimental results show that,compared with the traditional U-Net convolution neural network detection algorithm of pulmonary nodules,the improved algorithm improves the detection accuracy of small nodules by nearly 20%,and achieves more accurate localization of pulmonary lesions.
-
Computer Engineering. 2019, 45(4): 262-266,274. https://doi.org/10.19678/j.issn.1000-3428.0052533
The High Dynamic Range(HDR) imaging method based on Low-rank Matrix Completion(LRMC) removes the ghost image of the moving target in the resulting image by restoring the background information of the moving target occlusion.However,since the model ignores the local spatial information of the image,the edge of the HDR image cannot be effectively restored.In order to solve this problem,the spatial segmentation smoothness of moving targets is taken as an additional constraint.A background restoration model combining LRMC and total variational constraints is proposed,and the numerical calculation method of the model is given.At the same time,the low rank and sparsity and segmentation smoothness of moving targets of low dynamic range background image is utilized.Experimental results show that the proposed method can improve edge processing ability,and recovers edges well compared with two existing LRMC-based HDR imaging methods.
-
Computer Engineering. 2019, 45(4): 267-274. https://doi.org/10.19678/j.issn.1000-3428.0050099
In order to solve the problem of low accuracy of traditional fire recognition methods in underground mine,a fire image recognition algorithm based on improved Fruit Fly Optimization Algorithm(FOA)-Support Vector Machine(SVM) is proposed.YCrCb color space is used to segment the captured image,and several fire eigenvalues are extracted from the image sequence according to the early fire image features.The improved FOA algorithm based on group fusion is used to search the optimal kernel parameters and penalty factors of SVM,and the extracted fire image eigenvalues are used as the input of SVM to classify the sample data.Experimental results show that the accuracy of this method is 97.2%,and its classification effect is significantly better than FOA method and Particle Swarm Optimization(PSO) algorithm.
-
Computer Engineering. 2019, 45(4): 275-280. https://doi.org/10.19678/j.issn.1000-3428.0050057
Considering that most of existing methods on news content extraction do not take into account the topics features in the text,noise texts which are formatted like the news content may be identified as news content.Therefore,this paper proposes an extraction method based on topic weight of wildcard node.After parsing an HTML into DOM tree,it can match the DOM tree with a wildcard tree,and then calculates the weights of wildcard nodes.The wildcard nodes with high weight can be distinguish as news content.Experimental results show that the proposed method can reduce the false recognition rate of edge noise text of Web news content,and extract news content with higher accuracy rate compared with traditional news extraction methods.
-
Computer Engineering. 2019, 45(4): 281-287. https://doi.org/10.19678/j.issn.1000-3428.0049070
Aiming at the problem of premature convergence in the path planning of mobile robot based on Quantum-behaved Particle Swarm Optimization(QPSO) algorithm,an improved QPSO algorithm based on aggregation factor and phase mutation strategy is proposed.According to the objective function,the fitness value of the particle is calculated.By introducing the improved aggregation degree factor into the compression expansion factor,the search phase is divided,the individual position is updated by the phased mutation strategy,and the performance of the algorithm is tested.Experimental results show that compared with the FE-PSO algorithm,this algorithm has higher convergence accuracy and better stability.
-
Computer Engineering. 2019, 45(4): 288-295,301. https://doi.org/10.19678/j.issn.1000-3428.0050313
Aiming at the problem that the translation effect of single translation model in Uyghur-Chinese machine translation is poor and the translation difference between multiple translation models is large,a system fusion method based on paraphrase information is proposed.By extracting the Chinese interpretation information,the Chinese translation hypothesis is word aligned,and the word alignment information is used to construct and decode the confusion network,and the fusion result of the Uyghur-Chinese machine translation system is obtained.Experimental results show that compared with the single translation system HPSTW,this method can effectively improve the translation quality.
-
Computer Engineering. 2019, 45(4): 296-301. https://doi.org/10.19678/j.issn.1000-3428.0050948
In order to alleviate the problem of degraded emotional classification performance caused by fluctuation and difference of day-to-day Electroencephalogram(EEG) signals,an EEG feature extraction algorithm for EEG emotion classification based on the combination of Data Space Adaptation(DSA) and Common Spatial Pattern(CSP) iteration is proposed.For 5 days of EEG data from 12 subjects,the DSA algorithm is first used to perform spatial linear transformation to minimize the difference between cross-day EEG signals.Then,the CSP algorithm is used to transform it into an optimal subspace to maximize the variance between two emotional classes.The power spectral density features and differential lateral and causal features are extracted.Experimental results show that the proposed algorithm improves the accuracy and stability of emotional classification on EEG signals.
-
Computer Engineering. 2019, 45(4): 302-306,310. https://doi.org/10.19678/j.issn.1000-3428.0050489
Aiming at the problem of high power consumption and vulnerability to motion artifacts for long-term heart rate monitoring using pulse signals,an algorithm for extracting heart rate from compressed sampled pulse signals is proposed.The Lomb-Scargle spectral analysis method is used to extract the spectrum from the compressed sampled pulse signal and the uncompressed acceleration signal.The least squares spectrum subtraction method is used to obtain the differential spectrum.On this basis,the heart rate estimation is obtained by spectral peak tracking.Experimental results show that compared with TROIKA algorithm,the algorithm can reduce power consumption,has strong anti-interference and better heart rate estimation performance at 25 times compression ratio.
-
Computer Engineering. 2019, 45(4): 307-310. https://doi.org/10.19678/j.issn.1000-3428.0052193
In large-scale monitoring systems,failure of monitoring points can lead to data loss,which affects the accuracy of data analysis results.Therefoer,a method for complementing water conservancy data against self-encoding is proposed.The self-encoder is used to construct the network to extract the data features of the monitoring points,and then compete with the trained discriminant network to finally complete the monitoring data to be repaired.The experimental results show that compared with the graph-based regularized local sub-representation method,the proposed method has higher accuracy and less mean square error,which can effectively reconstruct the monitoring data.
-
Computer Engineering. 2019, 45(4): 311-315,320. https://doi.org/10.19678/j.issn.1000-3428.0050023
Aiming at the problem that unmanned aerial vehicle path planning only applies to static scenes,an autonomous flight inspection method is proposed.The drone stores the flight experience in the structure of the Incremental Hierarchical Discriminant Regression(IHDR) tree.The current position vector is input to search the IHDR tree to obtain the flight control amount.The output control amount is adjusted according to the deviation between the current position and the desired position,and the inspection of the artificial target is realized.Experimental results show that compared with the IDHR method,the learning time of the method is shortened by about 12.2%,and the method has high accuracy,and is suitable for drone inspection.
-
Computer Engineering. 2019, 45(4): 316-320. https://doi.org/10.19678/j.issn.1000-3428.0051016
Most existing electronic commerce trust algorithms ignores the uncertainty of consumers’ subjective preference and dosn’t consider language evaluation information.To solve the problem,a multi-dimension integrated trust model is proposed.Multi-dimension trust feedback information is integrated into interval-valued intuitionistic fuzzy numbers.Based on Technique for Order Preference by Similarity to Ideal Solution (TOPSIS),the neartude between the interval-valued intuitionsic fuzzy numbers and the ideal intuitionsic fuzzy numbers is calculated,and the trust of service entities is ranked according to the neartude.To prevent the internet fraud,the penalty term is added into the model.The simulation results show that the decision-makers’ risk preference,the quantitative and qualitative evaluation information are taken into consideration in the model,and its anti-fraud ability is better than the interval-valued cloud model.
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