Table of Content

10 March 2017, Volume 37 Issue 3

Previous Issue    Next Issue

Fine-grained scheduling policy based on erasure code

LIAO Hui, XUE Guangtao, QIAN Shiyou, LI Minglu

2017, 37(3):  613-619.  DOI: 10.11772/j.issn.1001-9081.2017.03.613

Asbtract ( )   PDF (1208KB) ( )  

References | Related Articles | Metrics

Aiming at the problems of long data acquisition delay and unstable data download in cloud storage system, a scheduling scheme based on storage node load information and erasure code technique was proposed. Firstly, erasure code was utilized to improve the delay performance of data retrieving in cloud storage, and parallel threads were used to download multiple data copies simultaneously. Secondly, a lot of load information about storage nodes was analyzed to figure out which performance indicators would affect delay performance, and a new scheduling algorithm was proposed based on load information. Finally, the open-source project OpenStack was used to build a real cloud computing platform to test algorithm performance based on real user request tracing and erasure coding. A large number of experiments show that the proposed scheme not only can achieve 15% lower average delay but also reduce 40% volatility of delay compared with other scheduling policies. It proves that the scheduling policy can effectively improve delay performance and stability of data retrieving in real cloud computing platform, achieving a better user experience.

Key technologies of distributed data stream processing based on big data

CHEN Fumei, HAN Dezhi, BI Kun, DAI Yongtao

2017, 37(3):  620-627.  DOI: 10.11772/j.issn.1001-9081.2017.03.620

Asbtract ( )   PDF (1326KB) ( )  

References | Related Articles | Metrics

In the big data environment, the real-time processing requirement of data stream is high, and data calculations require persistence and high reliability. Distributed Data Stream Processing System (DDSPS) can solve the problem of data stream processing in big data environment. Besides, it has the advantages of scalability and fault-tolerance of distributed system, and also has high real-time processing capability. Four subsystems and their key technologies of the DDSPS based on big data were introduced in detail. The different technical schemes of each subsystem were discussed and compared. At the same time, an example of data stream processing system structure to detect Distributed Denial of Service (DDoS) attacks was introduced, which can provide the technical reference for data stream processing theory research and application technology development under big data environment.

Analysis of large-scale distributed machine learning systems: a case study on LDA

TANG Lizhe, FENG Dawei, LI Dongsheng, LI Rongchun, LIU Feng

2017, 37(3):  628-634.  DOI: 10.11772/j.issn.1001-9081.2017.03.628

Asbtract ( )   PDF (1169KB) ( )  

References | Related Articles | Metrics

Aiming at the problems of scalability, algorithm convergence performance and operational efficiency in building large-scale machine learning systems, the challenges of the large-scale sample, model and network communication to the machine learning system were analyzed and the solutions of the existing systems were also presented. Taking Latent Dirichlet Allocation (LDA) model as an example, by comparing three open source distributed LDA systems-Spark LDA, PLDA+ and LightLDA, the differences in system design, implementation and performance were analyzed in terms of system resource consumption, algorithm convergence performance and scalability. The experimental results show that the memory usage of LightLDA and PLDA+ is about half of Spark LDA, and the convergence speed is 4 to 5 times of Spark LDA in the face of small sample sets and models. In the case of large-scale sample sets and models, the network communication volume and system convergence time of LightLDA is much smaller than PLDA+ and SparkLDA, showing a good scalability. The model of "data parallelism+model parallelism" can effectively meet the challenge of large-scale sample and model. The mechanism of Stale Synchronous Parallel (SSP) model for parameters, local caching mechanism of model and sparse storage of parameter can reduce the network cost effectively and improve the system operation efficiency.

Candidate category search algorithm in deep level classification

ZHANG Zhonglin, LIU Shuchang, JIANG Fentao

2017, 37(3):  635-639.  DOI: 10.11772/j.issn.1001-9081.2017.03.635

Asbtract ( )   PDF (968KB) ( )  

References | Related Articles | Metrics

Aiming at the problem of low classification accuracy and slow processing speed in deep classification, a candidate category searching algorithm for text classification was proposed. Firstly, the search, classification of two-stage processing ideas were introduced, and the weighting of the category hierarchy was analyzed and feature was updated dynamically by combining with the structure characteristics of the category hierarchy tree and the related link between categories as well as other implicit domain knowledge. Meanwhile feature set with more classification judgment was built for each node of the category hierarchy tree. In addition, depth first search algorithm was used to reduce the search range and the pruning strategy with setting threshold was applied to search the best candidate category for classified text. Finally, the classical K Nearest Neighbor (KNN) classification algorithm and Support Vector Machine (SVM) classification algorithm were applied to classification test and contrast analysis on the basis of candidate classes. The experimental results show that the overall classification performance of the proposed algorithm is superior to the traditional classification algorithm, and the average F₁ value is about 6% higher than the heuristic search algorithm based on greedy strategy. The algorithm improves the classification accuracy of deep text classification significantly.

Improving feature selection and matrix recovery ability by CUR matrix decomposition

LEI Hengxin, LIU Jinglei

2017, 37(3):  640-646.  DOI: 10.11772/j.issn.1001-9081.2017.03.640

Asbtract ( )   PDF (1235KB) ( )  

References | Related Articles | Metrics

To solve the problem that users and products can not be accurately selected in large data sets, and the problem that user behavior preference can not be predicted accurately, a new method of CUR (Column Union Row) matrix decomposition was proposed. A small number of columns were selected from the original matrix to form the matrix C, and a small number of rows were selected to form the matrix R. Then, the matrix U was constructed by Orthogonal Rotation (QR) matrix decomposition. The matrixes C and R were feature matrixes of users and products respectively, which were composed of real data, and enabled to reflect the detailed characters of both users as well as products. In order to predict behavioral preferences of users accurately, the authors improved the CUR algorithm in this paper, endowing it with greater stability and accuracy in terms of matrix recovery. Lastly, the experiment based on real dataset (Netflix dataset) indicates that, compared with traditional singular value decomposition, principal component analysis and other matrix decomposition methods, the CUR matrix decomposition algorithm has higher accuracy as well as better interpretability in terms of feature selection, as for matrix recovery, the CUR matrix decomposition also shows superior stability and accuracy, with a preciseness of over 90%. The CUR matrix decomposition has a great application value in the recommender system and traffic flow prediction.

Partitioning and mapping algorithm for in-memory computing framework based on iterative filling

BIAN Chen, YU Jiong, XIU Weirong, YING Changtian, QIAN Yurong

2017, 37(3):  647-653.  DOI: 10.11772/j.issn.1001-9081.2017.03.647

Asbtract ( )   PDF (1133KB) ( )  

References | Related Articles | Metrics

Focusing on the issue that the only one Hash/Range partitioning strategy in Spark usually results in unbalanced data load at Reduce phase and increases job duration sharply, an Iterative Filling data Partitioning and Mapping algorithm (IFPM) which include several innovative approaches was proposed. First of all, according to the analysis of job execute scheme of Spark, the job efficiency model and partition mapping model were established, the definitions of job execute timespan and allocation incline degree were given. Moreover, the Extendible Partitioning Algorithm (EPA) and Iterative Mapping Algorithm (IMA) were proposed, which reserved partial data into extend region by one-to-many partition function at Map phase. Data in extended region would be mapped by extra iterative allocation until the approximate data distribution was obtained, and the adaptive mapping function was executed by awareness of calculated data size at Reduce phase to revise the unbalanced data load in original region allocation. Experimental results demonstrate that for any distribution of the data, IFPM promotes the rationality of data load allocation from Map phase to Reduce phase and optimize the job efficiency of in-memory computing framework.

Weibo users credibility evaluation based on user relationships

LI Fumin, TONG Lingling, DU Cuilan, LI Yangxi, ZHANG Yangsen

2017, 37(3):  654-659.  DOI: 10.11772/j.issn.1001-9081.2017.03.654

Asbtract ( )   PDF (972KB) ( )  

References | Related Articles | Metrics

With the deepening of Weibo research, credibility evaluation of Weibo users has become a research hotspot. Aiming at the problem of Weibo users' credibility evaluation, a user confidence analysis method based on association was proposed. Taking Sina Weibo as the research object, firstly, seven characteristics of the user from three aspects: user information, interactive information and behavior information were analyzed, and the user self-evaluation credibility was got by using Analytic Hierarchy Process (AHP). Then, by using the user self-evaluation as the base point, the user relationship network as the carrier, and the potential users' evaluation relationship among the users, was improved the PageRank algorithm, and the user credibility evaluation model called User-Rank was proposed. The proposed model was used to evaluate comprehensively credibility of users by other users in relational network. Experiments on large scale Weibo real data show that the proposed method can obtain good evaluation results of user credibility.

Application case of big data analysis-robustness of a trading model

QIN Xiongpai, CHEN Yueguo, WANG Bangguo

2017, 37(3):  660-667.  DOI: 10.11772/j.issn.1001-9081.2017.03.660

Asbtract ( )   PDF (1417KB) ( )  

References | Related Articles | Metrics

The robustness of a trading model means that the model's profitability curve is less volatile and does not fluctuate significantly. To solve the problem of robustness of an algorithmic trading model based on Support Vector Regression (SVR), several strategies to derive a unified trading model and a portfolio diversification method were proposed. Firstly, the algorithm trade model based on SVR was introduced. Then, based on the commonly used indicators, a number of derived indicators were constructed for short term forecasting of stock prices. The typical patterns of recent price movements, overbought/oversold market conditions, and divergence of market conditions were characterized by these indicators. These indicators were normalized and used to train the trading model so that the model can be generalized to different stocks. Finally, a portfolio diversification method was designed. In the portfolio, the correlation between various stocks, sometimes leads to great investment losses; because the price of the stock with strong correlation changes in the same direction. If the trading model doesn't predict the price trend correctly, then stop loss will be triggered, and these stocks will cause loss in a mutual accelerated manner. Stocks were clustered into different categories according to the similarity, and a diversified portfolio was formed by selecting a number of stocks from different clustered categories. The similarity of stocks, was defined as the similarity of the recent profit curves on different stocks by trading models.Experiments were carried out on the data of 900 stocks for 10 years. The experimental results show that the transaction model can obtain excess profit rate over time deposit, and the annualized profit rate is 8.06%. The maximum drawdown of the trading model was reduced from 13.23% to 5.32%, and the Sharp ratio increased from 81.23% to 88.79%. The volatility of the profit margin curve of the trading model decreased, which means that the robustness of the trading model was improved.

Soft-sensing modeling based on improved extreme learning machine

ZHOU Xin, WANG Guoyin, YU Hong

2017, 37(3):  668-672.  DOI: 10.11772/j.issn.1001-9081.2017.03.668

Asbtract ( )   PDF (801KB) ( )  

References | Related Articles | Metrics

Extreme Learning Machine (ELM) has become a new method in soft-sensing due to its good generalization and fast training speed. However, ELM often needs more hidden layer nodes and its generalization is reduced in the parameter modeling for aluminum electrolysis production process. To solve the problem, a soft-sensing model based on Improved Extreme Learning Machine (IELM) was proposed. Firstly, rough set theory was applied to reduce the unnecessary, unrelated or reductant input variables, reducing the complexity of ELM input. After analyzing the relationship between the input variables and output variables by partial correlation coefficient, the input data was divided into two parts, namely the positive part and the negative part. Then, the corresponding ELM model was built according to the two parts. Finally, the soft-sensing model of molecular ratio based on the improved ELM was built. The simulation experimental results show that the soft-sensing model based on the IELM has better generalization and stability.

Query performance and data migration for social network database with shard strategy based on clustering analysis

LIANG Shuang, ZHOU Lihua, YANG Peizhong

2017, 37(3):  673-679.  DOI: 10.11772/j.issn.1001-9081.2017.03.673

Asbtract ( )   PDF (1109KB) ( )  

References | Related Articles | Metrics

Social network data has a certain degree of aggregation, namely the similar users are more prone to the same behavior. According to the conventional horizontal database shard method, a large amount of time and connection loss were consumed in order to access a plurality of databases in turn when performing the information query of these events. In order to solve this problem, the database shard strategy based on clustering analysis was proposed. Through clustering the characteristic scalars of social network subjects, the main body with the high aggregation was divided into one or as possible libraries to improve the query efficiency of the events, and to give consideration to load balancing, large data migration and other issues. The experimental results show that for the mainstream social networking events, the performance improvement of the proposed strategy is up to 23.4% at most, and local optimal load balance and zero data migration are realized. In general, the database shard strategy based on clustering analysis of social network, has a considerable advantage on improving query efficiency, balance load balancing and large data migration feasibility over the traditional conventional horizontal database shard method of cutting library.

Spherical embedding algorithm based on Kullback-Leibler divergence and distances between nearest neighbor points

ZHANG Bianlan, LU Yonggang, ZHANG Haitao

2017, 37(3):  680-683.  DOI: 10.11772/j.issn.1001-9081.2017.03.680

Asbtract ( )   PDF (773KB) ( )  

References | Related Articles | Metrics

Aiming at the problem that the existing spherical embedding algorithm cannot effectively embed the data into the low-dimensional space in the case that the distances between points far apart are inaccurate or absent, a new spherical embedding method was proposed, which can take the distances between the nearest neighbor points as input, and embeds high dimensional data of any scale onto the unit sphere, and then estimates the radius of the sphere which fit the distribution of the original data. Starting from a randomly generated spherical distribution, the Kullback-Leibler (KL) divergence was used to measure the difference of the normalized distance between each pair of neighboring points in the original space and the spherical space. Based on the difference, the objective function was constructed. Then, the stochastic gradient descent method with momentum was used to optimize the distribution of the points on the sphere until the result is stable. To test the algorithm, two types of spherical distribution data sets were simulated: which are spherical uniform distribution and Kent distribution on the unit sphere. The experimental results show that, for the uniformly distributed data, the data can be accurately embedded in the spherical space even if the number of neighbors is very small, the Root Mean Square Error (RMSE) of the embedded data distribution and the original data distribution is less than 0.00001, and the spherical radius of the estimated error is less than 0.000001; for spherical normal distribution data, the data can be embedded into the spherical space accurately when the number of neighbors is large. Therefore, in the case that the distance between points far apart are absent, the proposed method can still be quite accurate for low-dimensional data embedding, which is very helpful for the visualization of data.

Salient target detection algorithm based on contrast optimized manifold ranking

XIE Chang, ZHU Hengliang, LIN Xiao, MA Lizhuang

2017, 37(3):  684-690.  DOI: 10.11772/j.issn.1001-9081.2017.03.684

Asbtract ( )   PDF (1190KB) ( )  

References | Related Articles | Metrics

The existing boundary prior based saliency algorithm model has the problem of improper selection of reasonable saliency prior region, which leads to the inaccurate foreground region and influence the final result. Aiming at this problem, a salient target detection algorithm based on contrast optimized manifold ranking was proposed. The image boundary information was utilized to find the background prior. An algorithm for measuring the priori quality was designed by using three indexes, namely, saliency expection, local contrast and global contrast. A priori quality design with weighted addition replaced simple multiplication fusion to make the saliency prior more accurate. When the salient regions were extracted from the a priori, the strategy of selecting the threshold was changed, the foreground region was selected more rationally, and the saliency map was obtained by using the manifold ranking, so that the saliency detection result was more accurate. The experimental results show that the proposed algorithm outperforms the similar algorithms, reduces the noise, which is more suitable for human visual perception, and ahead of the depth learning method in processing time.

Design of DMA controller for multi-channel transmission system based on PCIe

LI Shenglan, JIANG Hongxu, FU Weijian, CHEN Jiao

2017, 37(3):  691-694.  DOI: 10.11772/j.issn.1001-9081.2017.03.691

Asbtract ( )   PDF (800KB) ( )  

References | Related Articles | Metrics

To reduce the impact of Programmed I/O (PIO) write latency in PCI express (PCIe) transmission process, too many times of interaction between the host and the embedded processing system and other issues on transmission bandwidth, a Direct Memory Access (DMA) controller based on command buffering mechanism was designed to improve the transmission bandwidth utilization. Using the internal command buffer of the Field-Programmable Gate Array (FPGA), the DMA controller could cache the data transfer request of the PC. The FPGA could dynamically access the storage space of the PC according to its own requirements and enhance the transmission flexibility. At the same time, a dynamic mosaic DMA scheduling method was proposed to reduce the times of host-to-hardware interaction and interrupt generation by merging the access requests of adjacent storage areas. In the system transmission rate test, the maximum write speed of DMA was 1631 MB/s, the maximum rate of DMA read was up to 1582 MB/s, the maximum of bandwidth was up to 85.4% of the theoretical bandwidth of PCIe bus. Compared with the traditional PIO mode DMA transfer method, DMA read bandwidth increased by 58%, DMA write bandwidth increased by 36%. The experimental results show that the proposed design can effectively improve the DMA transfer efficiency, and is significantly better than PIO method.

Objective quality assessment method of high dynamic range image

GUAN Feifan, YU Mei, SONG Yang, SHAO Hua, JIANG Gangyi

2017, 37(3):  695-698.  DOI: 10.11772/j.issn.1001-9081.2017.03.695

Asbtract ( )   PDF (813KB) ( )  

References | Related Articles | Metrics

Aiming at the problem that High Dynamic Range (HDR) image quality evaluation method does not consider the color and structure information of HDR image, a novel objective quality assessment method of HDR image was proposed. Firstly, the feature of visual fidelity about brightness and contrast was obtained based on the visual model of HDR-VDP-2.2. Then, the HDR image was transformed into the YIQ color space, and the color similarity and structural correlation coefficient were gotten by dealing with the Y, I, Q channel, respectively. Finally, Support Vector Regression (SVR) was used to fuse the features, and the objective evaluation value of the high dynamic range image quality could be obtained by predicting the similarity degree and the structural relevance degree. The experimental results show that compared with HDR-VDP-2.2, the Pearson correlation coefficient and Spearman rank correlation coefficient of the proposed method are increased by 23.09% and 25.34%, respectively; the Root Mean Square Error (RMSE) is reduced by 38.01%. The proposed method has higher consistency with subjective visual perception.

Diabetic retinal image classification method based on deep neural network

DING Pengli, LI Qingyong, ZHANG Zhen, LI Feng

2017, 37(3):  699-704.  DOI: 10.11772/j.issn.1001-9081.2017.03.699

Asbtract ( )   PDF (1070KB) ( )  

References | Related Articles | Metrics

Aiming at the problems of complex retinal image processing, poor generalization and lack of complete automatic recognition system, a complete retinal image automatic recognition system based on deep neural network was proposed. Firstly, the image was denoised, normalized, and data preprocessed. Then, a compact neural network model named CompactNet was designed. The structure parameters of CompactNet were inherited from AlexNet. The deep network parameters were adjusted adaptively based on the training data. Finally, the performance experiments were conducted on different training methods and various network structures. The experimental results demonstrate that the fine-tuning method of CompactNet is better than the traditional network training method, the classification index can reach 0.87, 0.27 higher than the traditional direct training. By comparing LeNet, AlexNet and CompactNet, CompactNet network model has the highest classification accuracy, and the necessity of preprocessing methods such as data amplification is confirmed by experiments.

User classification method based on multiple-layer network traffic analysis

MU Tao, CHEN Wei, CHEN Songjian

2017, 37(3):  705-710.  DOI: 10.11772/j.issn.1001-9081.2017.03.705

Asbtract ( )   PDF (1121KB) ( )  

References | Related Articles | Metrics

Accurate classification of users plays an important role in improving the quality of customized services, but for privacy considerations users, often do not meet the network service providers, refusing to provide personal information, such as location information, hobbies and so on. To solve this problem, by analyzing the multi-layer network traffic such as network layer and application layer under the premise of protecting user privacy, and then using machine learning methods such as K-means clustering and random forest algorithm to predict the user's geographic location types (such as apartments, campuses, etc.) and hobbies, and the relationship between geographic location types and the user interests was analyzed to improve the accuracy of user classification. The experimental results show that the proposed scheme can adaptively partition the user types and geographic location types, and improve the accuracy of user behavior analysis by correlating the user's geographic location type and the user type.

Device to device time division scheduling algorithm based on fairness

ZHAN Jinzhen, GUO Dawei, HUA Weixin

2017, 37(3):  711-716.  DOI: 10.11772/j.issn.1001-9081.2017.03.711

Asbtract ( )   PDF (936KB) ( )  

References | Related Articles | Metrics

To solve the problem of throughput degradation caused by slot scheduling delay and channel gain variation in Device to Device (D2D) communication resource allocation, a Fairness Time Division Scheduling (FTDS) algorithm was proposed. Firstly, the system model was established based on the spectrum reuse mode, and was transformed to a combinatorial optimization problem. Then, in the sub-optimal solution of the model, the scheduling period was divided into several equal-length slots by the FTDS algorithm, and the D2D users were assigned to different time divisions according to the priority policy, for the application scenario that D2D users are more than cellular users. Meanwhile, to balance Quality of Service (QoS) and system throughput, a satisfaction weight was constructed to restrict transmission rate, jointly determined the user scheduling priority. In the simulation, compared with TDS and RANDOM algorithm, the average throughput increase of FTDS algorithm was 11.09% and 40.64% respectively, and cumulative distribution of D2D scheduling frequency was more centralized by FTDS algorithm; meanwhile, the time delay of FTDS algorithm decreased as much as 31.22% compared with TDS.The simulation results show that FTDS algorithm has better throughput performance, more fair scheduling mechanism and smaller scheduling time delay.

Dynamic load balancing algorithm based on monitoring and adjusting of multiple detection engines

YANG Zhongming, LIANG Benlai, QIN Yong, CAI Zhaoquan

2017, 37(3):  717-721.  DOI: 10.11772/j.issn.1001-9081.2017.03.717

Asbtract ( )   PDF (794KB) ( )  

References | Related Articles | Metrics

To solve the load balance problem of multi-engine intrusion detection system, a dynamic load regulation algorithm of detection engine was proposed. Firstly, load was calculated by monitoring each engine node. Then, the scheduling of the heavy load node was performed by scheduling the overload or no-load node as a scheduling opportunity, and the nodes were traversed to adjust the load balancing. As the session for the scheduling unit, the algorithm was not the absolute average load for the purpose, just to ensure that the engine node does not appear overload or no load to achieve the basic goal. The KDD cup99 data set was used to simulate experiment. The experimental results show that compared with average load allocation algorithm and secure load allocation, the proposed algorithm has a significant effect on session-based load balancing, the running cost is lower, and the packet loss rate under heavy load are lower, which improves the detection rate of intrusion detection system.

Ant colony algorithm with gradient descent for solving multi-constrained quality of service routing

LIANG Benlai, YANG Zhongming, QIN Yong, CAI Zhaoquan

2017, 37(3):  722-729.  DOI: 10.11772/j.issn.1001-9081.2017.03.722

Asbtract ( )   PDF (1256KB) ( )  

References | Related Articles | Metrics

To solve the problem that many improved ant colony algorithms are not efficient to solve the problem of multi-constrained Quality of Service Routing (QoSR), such as slow convergence and local optimization, an Ant Colony Algorithm with Gradient Descent (ACAGD) was proposed. The gradient descent method was introduced into the local search of ant colony, and combined with residual pheromone, the next-hop selection strategy of ants was synthetically determined. Ant colony not only search for the next hop according to the pheromone concentration with certain probability, but also search for the next hop according to the gradient descent method with certain probability, which reduced the possibility that the traditional ant colony algorithm was easy to fall into the local optimum. The Waxman network model was used to randomly generate the network topology with different number of routing nodes. The experimental results show that compared with other improved ACO algorithms, the ACAGD can obtain the route with relatively low comprehensive cost while the convergence rate is not affected, and the stability of the algorithm is better.

Power allocation algorithm for two-tier underwater wireless sensor network using Stackelberg game

LI Xinbin, WANG Bei, HAN Song

2017, 37(3):  730-735.  DOI: 10.11772/j.issn.1001-9081.2017.03.730

Asbtract ( )   PDF (944KB) ( )  

References | Related Articles | Metrics

Focused on the issue that energy consumption is excessively high in the underwater wireless sensor cooperative communication networks, to balance the energy consumption between the nodes and increase the channel capacity of the system, a distributed power allocation game-theoretic algorithm based on the node residual energy was proposed. The trading model between the user node and the relay node was constructed as a two-tier Stackelberg game, so that the node with less residual energy could provide less power for forwarding service, otherwise the node with more residual energy could provide more power to service, so as to balance the energy consumption between nodes. Compared with the algorithm without considering the residual energy, the channel capacity increases by 9.4%, 23.1% and 16.7% when there are 2, 3, 4 relay nodes respectively. The simulation results show that the algorithm not only improves the total channel capacity of the system, but also prolongs the lifetime of the underwater sensor cooperative communication network.

Radar signal design based on nonlinear combination modulation of piecewise fitting

ZHANG Zhaoxia, LIU Jie, ZHAO Yan, HU Xiu, YANG Lingzhen

2017, 37(3):  736-740.  DOI: 10.11772/j.issn.1001-9081.2017.03.736

Asbtract ( )   PDF (735KB) ( )  

References | Related Articles | Metrics

The combinational modulated radar signal by linear frequency modulation and pseudo random code can increase signal complexity and reduce the probability of signal interception. However, this combination keeps the disadvantage of high sidelobe of linear frequency modulation and cannot overcome the increase of main lobe width caused by adding window function. Using the advantages of nonlinear frequency modulation signal generated by traditional stationary phase principle in suppressing sidelobe, a method which combined piecewise fitting nonlinear frequency modulation with Barker code modulation was proposed, and by using their respective advantages, the proposed method can not only reduce the ambiguity between distance and velocity, but also suppress the sidelobe. Finally, the simulation curve of explicit function model of this signal and ambiguity function were analyzed. The simulation results show that this method not only further reduces the ambiguity between distance and velocity and improves the performance of low probability signal interception, but also produces lower sidelobe than that of traditional nonlinear frequency modulation, which illustrates the feasibility and efficiency of the method used for radar signal.

Transmission channel calibration algorithm for digital instrument landing system

FENG Xiang, ZHANG Bin

2017, 37(3):  741-745.  DOI: 10.11772/j.issn.1001-9081.2017.03.741

Asbtract ( )   PDF (780KB) ( )  

References | Related Articles | Metrics

In the digital Instrumentation Landing System (ILS) transformation, aiming at the problem that the ILS adopts the amplitude angle measuring system and is sensitive to the distortion of the amplitude of the transmitted signal, a calibration algorithm of the transmission channel of the digital ILS was proposed. Firstly, the mathematical model of the transmitter's landing system was established, and the effect of the non-linearity of the transmission channel on the angular performance of the ILS was simulated. Secondly, a transmitter structure with a feedback loop in the digital instrumentation landing system was proposed. Finally, the transmission channel was calibrated by solving the inverse model of the transmission channel in the baseband using the Least Mean Square (LMS) algorithm, and using the inverse model to compensate the nonlinear distortion of the transmission channel. The simulation results show that the proposed algorithm can quickly estimate the inverse model of the transmission channel under noise conditions and has good calibration performance.

Fast learning algorithm of multi-output support vector regression with data-dependent kernel

WANG Dingcheng, ZHAO Youzhi, CHEN Beijing, LU Yiyi

2017, 37(3):  746-749.  DOI: 10.11772/j.issn.1001-9081.2017.03.746

Asbtract ( )   PDF (735KB) ( )  

References | Related Articles | Metrics

For the Multi-output Support Vector Regression (MSVR) algorithm based on gradient descent method in the process of model parameter fitting, the convergence rate is slow and the prediction accuracy is low. A modified version of the Quasi-Newton algorithm (BFGS) with second-order convergence rate based on the rank-2 correction rule was used to fit the model parameters of MSVR algorithm. At the same time, to ensure the decrease of the iterative process and the global convergence, the step size factor was determined by the non-exact linear search technique. Based on the analysis of the geometry structure of kernel function in Support Vector Machine (SVM), a data-dependent kernel function was substituted for the traditional kernel function, and the multi-output data-dependent kernel support vector regression model was generated. The model was compared with the multi-output support vector regression model based on gradient descent method and modified Newton method. The experimental results show that in the case of 200 samples, the iterative time of the proposed algorithm is 72.98 s, the iterative time of modified Newton's algorithm is 116.34 s and the iterative time of gradient descent method is 2065.22 s. The proposed algorithm can reduce the model iteration time and has faster convergence speed.

Particle swarm optimization algorithm based on self-adaptive excellence coefficients for solving traveling salesman problem

CHENG Biyun, LU Haiyan, HUANG Yang, XU Kaibo

2017, 37(3):  750-754.  DOI: 10.11772/j.issn.1001-9081.2017.03.750

Asbtract ( )   PDF (988KB) ( )  

References | Related Articles | Metrics

To solve the problem that basic discrete Particle Swarm Optimization (PSO) algorithm often leads the computation process into local optimum and premature convergence when applied to Traveling Salesman Problem (TSP), a PSO based on Self-adaptive Excellence Coefficients (SECPSO) algorithm was proposed. To improve the global search ability, heuristic information was further utilized to modify the static excellence coefficients of paths based on previous work, so that these coefficients could be adjusted adaptively and dynamically according to the process of searching for the solutions. Furthermore, a 3-opt search mechanism was added to improve the accuracy of the solution and the convergence rate of the algorithm. Through simulation experiments with Matlab, the performance of the proposed algorithm was evaluated using several classical examples in the international general TSP database (TSPLIB). The experimental results indicate that the proposed SECPSO algorithm performs better in terms of global search ability and convergence rate compared with several other algorithms, and thus is a potential intelligent algorithm for solving TSP.

Virtual network embedding algorithm based on multi-objective particle swarm optimization

LI Zhen, ZHENG Xiangwei, ZHANG Hui

2017, 37(3):  755-759.  DOI: 10.11772/j.issn.1001-9081.2017.03.755

Asbtract ( )   PDF (940KB) ( )  

References | Related Articles | Metrics

In virtual network mapping, most studies only consider one mapping object, which can not reflect the interests of many aspects. To solve this problem, a Virtual Network Embedding algorithm based on Multi-objective Particle Swarm Optimization (VNE-MOPSO) was proposed by combining multi-objective algorithm and Particle Swarm Optimization (PSO) algorithm. Firstly, the crossover operator was introduced into the basic PSO algorithm to expand the search space of population optimization. Secondly, the non-dominated sorting and crowding distance sorting were introduced into the multi-objective optimization algorithm, which can speed up the population convergence. Finally, by minimizing both the cost and the node load balance degree as the virtual network mapping objective function, a multi-objective PSO algorithm was proposed to solve the Virtual Network Mapping Problem (VNMP). The experimental results show that the proposed algorithm can solve the VNMP, which has advantages in network request acceptance rate, average cost, average node load balance degree, and infrastructure provider's profit.

Multiple interactive artificial bee colony algorithm and its convergence analysis

LIN Kai, CHEN Guochu, ZHANG Xin

2017, 37(3):  760-765.  DOI: 10.11772/j.issn.1001-9081.2017.03.760

Asbtract ( )   PDF (893KB) ( )  

References | Related Articles | Metrics

Aiming at the shortcomings of Artificial Bee Colony (ABC) algorithm, which is not easy to jump out of the local optimal value, a Multiple Interactive Artificial Bee Colony (MIABC) algorithm was proposed. The proposed algorithm was based on the basic ABC algorithm, involved the random neighborhood search strategy and the cross-dimensional search strategy, and improved the treatment when bees exceed the limit, so the search way of the algorithm became various, the algorithm itself had stronger bound and it's hard to trap in the local optimal value. Meanwhile, the convergence analysis and performance test were carried out. The simulation result based on five kinds of classic benchmark functions and experimental results for time complexity show that comparing with the standard ABC algorithm and basic Particle Swarm Optimization (PSO), this proposed method has faster convergence speed which is increased by about 30% and 65% at 1E-2 accuracy and better search precision, besides, it has significant advantages in solving high dimensional problems.

Efficient and secure deduplication cloud storage scheme based on proof of ownership by Bloom filter

LIU Zhusong, YANG Zhangjie

2017, 37(3):  766-770.  DOI: 10.11772/j.issn.1001-9081.2017.03.766

Asbtract ( )   PDF (807KB) ( )  

References | Related Articles | Metrics

Convergent encryption algorithm is generally used in deduplication cloud storage system, the data can be encrypted by using the hash value as the encryption key, so that the same data is encrypted to obtain the same ciphertext, and the deletion of the duplicate data can be realized, then through the Proof of oWnership (PoW), the authenticity of user data can be verified to protect data security. Aiming at the problem that the time overhead of Proof of oWnership (PoW) is too high, which leads to the degradation of the whole system performance, an efficient security method based on Bloom Filter (BF) was proposed to verify the user hash value and the initialization value efficiently. Finally, a BF scheme supporting fine-grained data deduplication was proposed. When the file level data was duplicated, the PoW was needed; otherwise, only partial block level data duplication detection was needed. The simulation experiment results show that, the key space overhead of the proposed BF scheme is lower than the classical Baseline scheme, and the time cost of the BF scheme is also lower than the Baseline scheme; and with the increase of data size, the performance advantage of BF scheme is more obvious.

Mechanism of security situation element acquisition based on deep auto-encoder network

ZHU Jiang, MING Yue, WANG Sen

2017, 37(3):  771-776.  DOI: 10.11772/j.issn.1001-9081.2017.03.771

Asbtract ( )   PDF (941KB) ( )  

References | Related Articles | Metrics

To reduce the time complexity of situational element acquisition and cope with the low detection accuracy of small class samples caused by imbalanced class distribution of attack samples in large-scale networks, a situation element extraction mechanism based on deep auto-encoder network was proposed. In this mechanism, the improved deep auto-encoder network was introduced as basic classifier to identify data type. On the one hand, in the training of the auto-encoder network, the training rule based on Cross Entropy (CE) function and Back Propagation (BP) algorithm was adopted to overcome the shortcoming of slow weights updating by the traditional variance cost function. On the other hand, in the stage of fine-tuning and classification of the deep network, an Active Online Sampling (AOS) algorithm was applied in the classifier to select the samples online for updating the network weights, so as to eliminate redundancy of the total samples, balance the amounts of all sample types, improve the classification accuracy of small class samples. Simulation and analysis results show that the proposed scheme has a good accuracy of situation element extraction and small communication overhead of data transmission.

Extracting kernel basis using differential evolution algorithm for packet matching

WANG Zelin, HAO Shuixia

2017, 37(3):  777-781.  DOI: 10.11772/j.issn.1001-9081.2017.03.777

Asbtract ( )   PDF (797KB) ( )  

References | Related Articles | Metrics

Aiming at the speed of packet matching in network firewall, router and other equipment, a differential evolution algorithm was proposed to extract the multi-layer core base of package matching. The multi-layer foundation was used to describe the multi-layer characteristics of the packet. In each layer, the bit basics and entitative basics were extracted using differential evolution algorithm and average self-information and the average mutual information were used to evaluate the quality of kernel basis. This method was adapt to select the number of layers of the extracted entity base according to the actual size of rule base, which is very suitable for the growth of rule base. The experimental results show that The proposed algorithm is the first known algorithm to be applied to packet matching efficiently. Compared with RFC (Recursive Flow Classification) algorithm and RDEPM (Real-based Differential Evolution Packet Matching) algorithm, the performance of the proposed algorithm is superior in terms of time efficiency and space efficiency.

Network traffic classification based on Plane-Gaussian artificial neural network

YANG Xubing, FENG Zhe, GU Yifan, XUE Hui

2017, 37(3):  782-785.  DOI: 10.11772/j.issn.1001-9081.2017.03.782

Asbtract ( )   PDF (792KB) ( )  

References | Related Articles | Metrics

Aiming at the problems of network flow monitoring (classification) in complex network environment, a stochastic artificial neural network learning method was proposed to realize the direct classification of multiple classes and improve the training speed of learning methods. Using Plane-Gaussian (PG) artificial neural network model, the idea of stochastic projection was introduced, and the network connection matrix was obtained by calculating the pseudo-inverse analysis. Theoretically, it can be proved that the network has global approximation ability. The artificial simulation was carried out on artificial data and standard network flow monitoring data. Compared with the Extreme Learning Machine (ELM) and PG network using the random method, the analysis and experimental results show that: 1)the proposed method inherits the geometric characteristics of the PG network and is more effective for the planar distributed data; 2)it has comparable training speed to ELM, but significantly faster than PG network; 3)among the three methods, the proposed method is more suitable for solving the problem of network flow monitoring.

Modal parameter identification of vibration signal based on unsupervised learning convolutional neural network

FANG Ning, ZHOU Yu, YE Qingwei, LI Yugang

2017, 37(3):  786-790.  DOI: 10.11772/j.issn.1001-9081.2017.03.786

Asbtract ( )   PDF (905KB) ( )  

References | Related Articles | Metrics

Aiming at the problem that most of the existing time-domain modal parameter identification methods are difficult to set order and resist noise poorly, an unsupervised learning Convolution Neural Network (CNN) method for vibration signal modal identification was proposed. The proposed algorithm was improved on the basis of CNN. Firstly, the CNN applied to two-dimensional image processing was changed into the CNN to deal with one-dimensional signal. The input layer was changed into the vibration signal set of modal parameters to be extracted, and the intermediate layer was changed into several one-dimensional convolution layers, sampled layers, and output layer was the set of N-order modal parameters corresponding to the signal. Then, in the error evaluation, the network calculation result (N-order modal parameter set) was reconstructed by the vibration signals. Finally, the squared sum of the difference between the reconstructed signal and the input signal was taken as the network learning error, which makes the network become an unsupervised learning network, and avoids the ordering problem of modal parameter extraction algorithm. The experimental results show that when the constructed CNN is applied to modal parameter extraction, compared with the Stochastic Subspace Identification (SSI) algorithm and its Local Linear Embedding (LLE) algorithm, the convolutional neural network identification accuracy is higher than that of the SSI algorithm and the LLE algorithm under noise interference. It has strong noise resistance and avoids the ordering problem.

Social recommendation algorithm combining rating and trust relation

HU Yun, LI Hui, SHI Jun

2017, 37(3):  791-795.  DOI: 10.11772/j.issn.1001-9081.2017.03.791

Asbtract ( )   PDF (814KB) ( )  

References | Related Articles | Metrics

To solve the problem of data sparsity and cold start which is prevalent in recommender system, a new social recommendation algorithm was proposed, which integrates rating and trust relation. Firstly, the initial trust value of the new user in the network was reasonably assigned, which solves the problem of cold start of the new user. Since the user's preferences were affected by his friends, the user's own feature vector was modified by the trust matrix between friends, which solves the problem of user's feature vector construction and trust transition. The experimental results show that the proposed algorithm has a significant performance improvement over the traditional social network recommendation algorithm.

Ordered decision-making based on preference inconsistence-based entropy

PAN Wei, SHE Kun

2017, 37(3):  796-800.  DOI: 10.11772/j.issn.1001-9081.2017.03.796

Asbtract ( )   PDF (814KB) ( )  

References | Related Articles | Metrics

Aiming at the problem of preference decision in multi-rule ordered decision-making system, according to the preference inconsistency of ordered decision-making, a preference decision-making method based on preference inconsistent entropy was proposed. Firstly, the Preference Inconsistence Entropy of Object (PIEO) was defined and used to measure the degree of preference inconsistency for a particular sample relative to the sample set. Then, according to that different attributes have different importances to the preference decision, a weighted Preference Inconsistence-based Entropy of Object (wPIEO) was proposed. Moreover, combining wPIEO with attribute preference inconsistency entropy in measuring attribute importance, a weighting method based on attribute preference inconsistent entropy was proposed. Finally, a preference decision algorithm based on sample preference inconsistent entropy was proposed. Two data sets, Pasture Production and Squalsh, were used to simulate the experiment. After the global Preference Inconsistent Entropy (gPIE) classification, the preference inconsistent entropy of each attribute was generally smaller than the entropy value based on the preference inconsistent entropy classification based on the up and down preferences, and it was closer to the preference inconsistent entropy of the original decision, which indicates that the classification based on gPIE was better than the other two cases. The classification deviation was as low as 0.1282, indicating that the classification results are close to the original decision.

Quantitative detection of face location in videos

WEI Wei, MA Rui, WANG Xiaofang

2017, 37(3):  801-805.  DOI: 10.11772/j.issn.1001-9081.2017.03.801

Asbtract ( )   PDF (869KB) ( )  

References | Related Articles | Metrics

Available face detection and evaluation standards are usually only a qualitative detection of the face existing, and have no strict norms for the quantitative description of the face location in videos.In addition, some researches such as video face replacement have higher requirements for the continuity of the face position in the video sequences. To solve these two problems, compared with the previous face detection algorithms and the face tracking evaluation standards, a quantitative detection standard of the human face position in the video was proposed, and a modified method of video face position detection was put forward. The initial face location was firstly detected in the target area by the improved Haar-Like cascade classifier; then the pyramid optical flow method was used to predict the position of the face, at the same time the forward-backward error detection mechanism was introduced to the self-checking of results, and finally the location of human face was determined. The experimental results show that the detection standard can give the evaluation of the quantitative description of the detection algorithm in the video face detection, and the proposed detection algorithm has a great improvement in the time consistency of face position in the detection results.

Tracking method of multi-resolution LK optical flow combined with SURF

LI Dan, BAO Rong, SUN Jinping, XIAO Liqing, DANG Xiangying

2017, 37(3):  806-810.  DOI: 10.11772/j.issn.1001-9081.2017.03.806

Asbtract ( )   PDF (1022KB) ( )  

References | Related Articles | Metrics

Aiming at the problem of tracking instability of the Lucas-Kanade (LK) algorithm for the complex situation of moving target deformation, fog and haze, high-speed, uneven illumination and partial occlusion in traffic monitoring, a tracking algorithm based on multi-resolution LK optical flow algorithm and Speed Up Robust Features (SURF) was proposed. The problem tracking failure for large-scale motion between frames of same pixel point in the traditional LK algorithm was solved by the proposed method, and the SURF scale invariant feature transformation algorithm was combined, feature points for optical flow tracking were extracted, and an adaptive template real-time update strategy was developed; the amount of optical flow calculation was reduced while enhancing the resistance ability of moving targets against complex environments. The experimental results show that the feature points matching of the new method is accurate and fast, which has strong adaptability and it is stable in the complicated traffic environment.

Scale adaptive improvement of kernel correlation filter tracking algorithm

QIAN Tanghui, LUO Zhiqing, LI Guojia, LI Yingyun, LI Xiankai

2017, 37(3):  811-816.  DOI: 10.11772/j.issn.1001-9081.2017.03.811

Asbtract ( )   PDF (961KB) ( )  

References | Related Articles | Metrics

To solve the problem that Circulant Structure of tracking-by-detection with Kernels (CSK) is difficult to adapt to the target scale change, a multi-scale kernel correlation filter classifier was proposed to realize the scale adaptive target tracking. Firstly, the multi-scale image was used to construct the sample set, the multi-scale kernel correlation filtering classifier was trained by the sample set, for target size estimation to achieve the goal of the optimal scale detection, and then the samples collected on the optimal target scale were used to update the classifier on-line to achieve the scale-adaptive target tracking. The comparative experiments and analysis illustrate that the proposed algorithm can adapt to the scale change of the target in the tracking process, the error of the eccentricity is reduced to 1/5 to 1/3 that of CSK algorithm, which can meet the needs of long time tracking in complex scenes.

Self-examples reconstruction of city street image from driving recorder

YANG Wei, XIE Weicheng, JIANG Wenbo, SHI Linyu

2017, 37(3):  817-822.  DOI: 10.11772/j.issn.1001-9081.2017.03.817

Asbtract ( )   PDF (1058KB) ( )  

References | Related Articles | Metrics

In order to ensure the high speed of image display and storage in real-time, the image captured by the popular driving recorder usually shows a low resolution, which has a serious impact on effective image information acquisition under unexpected situation. To solve this problem, a perspective transformation based on self-examples of the images and high-frequency compensation were used to reconstruct the city street images with low resolution. Perspective transformation was added to the affine transformation to match image patches, match image patch and high frequency compensation was used to recover the lost high frequency information of each matched image patch when image pyramid was constructed. The image pyramid was searched by non-local multi-scale method to get the matched patches, which were synthesized to obtain the images of high resolution. Many low resolution street view images were used to verify the effectiveness of this algorithm. Compared it to existing typical algorithms such as ScSR (Sparse coding Super-Resolution), Upscaling, SCN (Sparse Coding based Network), the experimental results show that the algorithm in several blind evaluation indices is better than other algorithms and it can improve the image resolution while keeping the edges and details of the image.

Application of improved spatially constrained Bayesian network model to image segmentation

ZHANG Haiyan, GAO Shangbing

2017, 37(3):  823-826.  DOI: 10.11772/j.issn.1001-9081.2017.03.823

Asbtract ( )   PDF (809KB) ( )  

References | Related Articles | Metrics

Aiming at the problem of iterative convergence of Markov chain Monte Carlo method, an improved spatially constrained Bayesian network model was proposed and applied in the image segmentation domain based on the Gaussian mixture model with spatial smoothing constraint. Latent Dirichlet Allocation (LDA) probability density model and the parameter mix process of Gauss-Markov theorem were used to achieve parameter smoothing. According to the spatial information transcendental transformation operation, the LDA conformance polynomial distribution was introduced into the context hybrid structure of the pixel to be used to replace the mapping operation in the traditional expectation maximization algorithm. LDA parameters were represented by a closed form, which facilitated to accurately estimate the relative proportion of MAP (Maximum A Posteriori) framework to context mixture structure. The experimental results in terms of PRI (Probabilistic Rand Index), VoI (Variation of Information), GCE (Global Consistency Error) and BDE (Boundary Displacement Error) show that the proposed method has better effect in image segmentation, its robustness is less influenced by Gauss noise compared with JSEG (Joint Systems Engineering Group), CTM (Current Transformation Matrix) and MM (Maximum A Posteriori Probability-Maximum Likelihood).

Binarization method with local threshold based on image blocks

ZHANG Jieyu

2017, 37(3):  827-831.  DOI: 10.11772/j.issn.1001-9081.2017.03.827

Asbtract ( )   PDF (889KB) ( )  

References | Related Articles | Metrics

Aiming at the defects of local threshold binarization methods resulting in false or broken targets, a local threshold binarization method based on image blocks was proposed. Firstly, the image was divided into several sub-blocks and the distribution of gray-value in each block was analyzed. Then, a local window of a certain size was moved within the image and the gray-value variation of the pixels in this local window was compared with that in a larger area including aforementioned local window. The larger area consists of all the sub-blocks currently covered by the window template to determine whether the window is gray-value flat (or violent). Finally, a specific binarization scheme was given according to different regions. Seven different algorithms were used to binarize four different types of four sets of images. The experimental results show that the proposed algorithm has the best performance in masking the background noise and preserving the target details. In particular, the algorithm can get the highest recall rate and accuracy rate through quantitative analysis of the license plate image binarization results.

Adaptive four-dot midpoint filter for removing high density salt-and-pepper noise in images

ZHANG Xinming, KANG Qiang, CHENG Jinfeng, TU Qiang

2017, 37(3):  832-838.  DOI: 10.11772/j.issn.1001-9081.2017.03.832

Asbtract ( )   PDF (1209KB) ( )  

References | Related Articles | Metrics

In view of poor denoising performance and unideal speed of the current median filter, a fast and Adaptive Four-dot Midpoint Filter (AFMF) was proposed. Firstly, noise pixels and non-noise pixels of an image were identified using a simple extreme method to reduce the computational complexity. Then, the traditional full-point window was discarded, instead of median filtering, but on the basis of switch filtering and clipping filtering, a new nonlinear filtering method named midpoint filtering was adopted to simplify the algorithm flow, improve the calculation efficiency, improve the denoising effect. Finally, starting from a 3×3 window from inside to outside, the window was gradually enlarged to form adaptive filtering, until all the noise pixels were processed, the setting of window size parameters was avoided. The experimental results show that compared with AMF, SAMF, MDBUTMF and DBCWMF, AFMF not only has better denoising performance but also faster operation speed (about 0.18 s), but also does not need to set parameters, which is easy to operate and has strong practicability.

Mountain altitude extraction from single remote sensing image based on dark channel prior

SHENG Tingting, CHEN Qiang, SUN Quansen

2017, 37(3):  839-843.  DOI: 10.11772/j.issn.1001-9081.2017.03.839

Asbtract ( )   PDF (901KB) ( )  

References | Related Articles | Metrics

The altitude information extracted from a single remote sensing image can be applied to detect the natural disaster, such as landslide or mud-rock flow. An approach based on dark channel prior was proposed for the altitude extraction from a single remote sensing image, which considers the influence of shadow. The approach was based on dark channel prior, and meanwhile a solution to overcome the effect of mountain shadow was given. The quantitative and qualitative analysis of a large number of mountain remote sensing images demonstrates that the proposed algorithm can obtain the accurate relative altitude information. In conclusion, the improved algorithm is effective for the extraction of the relative altitude from single remote sensing image of mountain with shadows.

Tooth segmentation algorithm based on segmentation of feature line

XIAO Bing, WEI Xin, HU Wei, XIA Hongjian

2017, 37(3):  844-848.  DOI: 10.11772/j.issn.1001-9081.2017.03.844

Asbtract ( )   PDF (916KB) ( )  

References | Related Articles | Metrics

Tooth segmentation plays an important role in computer-aided orthodontics. However, many published approaches directly separate teeth from dental mesh without dealing with the region fusion, which leads to inaccurate results and incomplete segmented teeth with side shape lacked. Meanwhile, existing tooth shape modeling schemes are interaction-intensive and inefficient. To resolve this problem, a new tooth segmentation approach based on segmentation of feature line was proposed. Feature region was selected according to mean curvature, and morphologic algorithm was used to extract dentition line. The fusion region was automatically recognized by the feature line segmenting and branch points matching algorithm as well as morphologic dilation. The restoration result was automatically obtained by repairing holes with matched branch points. After the gingival margin lines between adjacent teeth were extracted, the teeth were segmented by all the gingival margin lines. Experimental results demonstrate that the poposed approach is accurate, the segmented teeth have complete side feature. In addition, the approach avoids user interactions in the stage of tooth shape modeling, thus improving the whole efficience by 60%-90% compared with the method which manually identifies and removes the interdental adhesion area and reconstructs the missing tooth surface by surface energy constraint.

Automatic detection of blowholes defects in X-ray images of thick steel pipes

CHEN Benzhi, FANG Zhihong, XIA Yong, ZHANG Ling, LAN Shouren, WANG Lisheng

2017, 37(3):  849-853.  DOI: 10.11772/j.issn.1001-9081.2017.03.849

Asbtract ( )   PDF (866KB) ( )  

References | Related Articles | Metrics

Due to the intensity distribution of X-ray image of thick steel pipe is not uniform, the contrast is low, the noise is big, and the size, shape, position and contrast of the blowholes defects are different, it is difficult to detect various types of blowholes automatically. Aiming at the problems that the traditional defect detection algorithm has a large workload of manually marking defect data, and the edge of the weld is difficult to accurately extract and other issues, a new unsupervised learning algorithm was proposed for the detection of various blowholes defects. Firstly, fast Independent Component Analysis (ICA) was used to learn a set of independent base vectors from the steel pipe X-ray image set, and a linear combination of the base vectors was used to selectively reconstruct the test image with blowholes defect. Then, the test image was subtracted from its reconstructed image to obtain the difference image, and the various blowholes were separated from the difference image by global threshold. There were 320 images in the training set and 60 images in the test set. The average sensitivity and accuracy of the proposed algorithm were 90.5% and 99.7%. The experimental results show that the algorithm can accurately detect all kinds of blowholes defects without manual marking the data or extracting the edge of the weld.

Trajectory data clustering algorithm based on spatio-temporal pattern

SHI Lukui, ZHANG Yanru, ZHANG Xin

2017, 37(3):  854-859.  DOI: 10.11772/j.issn.1001-9081.2017.03.854

Asbtract ( )   PDF (1146KB) ( )  

References | Related Articles | Metrics

Because the existing trajectory clustering algorithms in the similarity measurement usually used the spatial characteristics as the standards the characteristics lacking the consideration of temporal, a trajectory data clustering algorithm based on spatial-temporal pattern was proposed. The proposed algorithm was based on partition-and-group framework. Firstly, the trajectory feature points were extracted by using the curve edge detection method. Then the sub-trajectory segments were divided according to the trajectory feature points. Finally, the clustering algorithm based on density was used according to the spatio-temporal similarity between sub-trajectory segments. The experimental results show that the trajectory feature points extracted using the proposed algorithm are more accurate to describe the trajectory structure under the premise that the feature points have better simplicity. At the same time, the similarity measurement based on spatio-temporal feature obtains better clustering result by taking into account both spatial and temporal characteristics of trajectory.

New spatio-temporal index method based on real-time data and query log distribution

MENG Xuechao, YE Shaozhen

2017, 37(3):  860-865.  DOI: 10.11772/j.issn.1001-9081.2017.03.860

Asbtract ( )   PDF (977KB) ( )  

References | Related Articles | Metrics

In the era of large data, the data has the characteristics of large volume, obvious spatio-temporal complexity, high real-time requirement, and etc. However, the traditional method of indexing large-scale spatio-temporal data based on tree structure has the problem of low utilization of storage space and low efficiency of query. In order to solve this problem, a new method named HDL-index was proposed to establish the spatio-temporal index based on the distribution of data and historical query records. On the one hand, the whole area was partitioned based on the spatial distribution of the data. On the other hand, taking into account the continuity of query, the query-models were obtained after density-based clustering on historical query objects, and then based on the model coordinates and query granularity of the overall query area segmentation, the two indexes were merged based on their GeoHash codes, and finally the optimal index coding was obtained. HDL-index takes better account of the data distribution and users' queries, making the index on the frequent query area more refined. Compared with the efficiency of the similar method, the efficiency of the index creation is improved by 50%, and the query efficiency of the hotspot region can be increased by more than 75% when the data is evenly distributed in the real aeronautical data set.

Online feature selection based on feature clustering ensemble technology

DU Zhenglin, LI Yun

2017, 37(3):  866-870.  DOI: 10.11772/j.issn.1001-9081.2017.03.866

Asbtract ( )   PDF (1000KB) ( )  

References | Related Articles | Metrics

According to the new application scenario with both historical data and stream features, an online feature selection based on group feature selection algorithm and streaming features was proposed. To compensate for the shortcomings of single clustering algorithm, the idea of clustering ensemble was introduced in the group feature selection of historical data. Firstly, a cluster set was obtained by multiple clustering using k-means method, and the final result was obtained by integrating hierarchical clustering algorithm in the integration stage. In the online feature selection phase of the stream feature data, the feature group generated by the group structure was updated by exploring the correlation among the features, and finally the feature subset was obtained by group transformation. The experimental results show that the proposed algorithm can effectively deal with the online feature selection problem in the new scenario, and has good classification performance.

Robust feature selection and classification algorithm based on partial least squares regression

SHANG Zhigang, DONG Yonghui, LI Mengmeng, LI Zhihui

2017, 37(3):  871-875.  DOI: 10.11772/j.issn.1001-9081.2017.03.871

Asbtract ( )   PDF (818KB) ( )  

References | Related Articles | Metrics

A Robust Feature Selection and Classification algorithm based on Partial Least Squares Regression (RFSC-PLSR) was proposed to solve the problem of redundancy and multi-collinearity between features in feature selection. Firstly, the consistency coefficient of sample class based on neighborhood estimation was defined. Then, the k Nearest Neighbor (kNN) operation was used to select the conservative samples with local class structure stability, and the partial least squares regression model was used to construct the robust feature selection. Finally, a partial least squares classification model was constructed using the class consistency coefficient and the preferred feature subset for all samples from a global structure perspective. Five data sets of different dimensions were selected from the UCI database for numerical experiments. The experimental results show that compared with four typical classifiers-Support Vector Machine (SVM), Naive Bayes (NB), Back-Propagation Neural Network (BPNN) and Logistic Regression (LR), RFSC-PLSR is more efficient in low-dimensional, medium-dimension, high-dimensional and other different cases, and shows stronger competitiveness in classification accuracy, robustness and computational efficiency.

Administrative division extracting algorithm for non-normalized Chinese addresses

LI Xiaolin, HUANG Shuang, LU Tao, LI Lin

2017, 37(3):  876-882.  DOI: 10.11772/j.issn.1001-9081.2017.03.876

Asbtract ( )   PDF (1226KB) ( )  

References | Related Articles | Metrics

Chinese addresses on the Internet are always non-normalized, which cannot be used directly in location-based services. To solve the problem, an algorithm to extract administrative divisions from non-normalized Chinese addresses was proposed. Firstly, preprocessing "road" feature word grouping for original data; using administrative division dictionary and moving window maximum matching algorithm, extract all possible administrative region data sets from Chinese address. Then, using the Chinese administrative divisions between the elements of the hierarchical relationship between the characteristics, the administrative set conditional set operation rule was established and the acquired data set was aggregated. using the administrative division of matching, a set of administrative division set rules were established to calculate the credibility of the administrative division. Finally, the credibility of the maximum amount of information the most complete Chinese address of the administrative divisions were obtained. By using the extracted from the Internet about 250000 Chinese address data whether the use of "road" feature word packet processing and whether to carry on the credibility calculation process was verified for the availability of the algorithm, and with the current address matching technology for comparison, the accuracy rate of 93.51%.

Dynamic path planning for autonomous driving with avoidance of obstacles

ZHOU Huizi, HU Xuemin, CHEN Long, TIAN Mei, XIONG Dou

2017, 37(3):  883-888.  DOI: 10.11772/j.issn.1001-9081.2017.03.883

Asbtract ( )   PDF (1001KB) ( )  

References | Related Articles | Metrics

To deal with the problem of dynamic path planning for autonomous driving with avoidance of obstacles, a real-time dynamic path planning approach was proposed to avoid obstacles in real-time under the condition of knowing initial vehicle position, speed, orientation and the obstacle positions. Firstly, a base frame of the road was constructed using the continuity of the second derivative for cubic spline curves combined with the information of the road edges and lanes. Secondly, the s-q coordinate system was established using the position and orientation of the vehicle and the curvature of the road. Then a set of smooth curves from the current position to the destination were generated as the path candidates in the s-q coordinate system. Finally, considering the factors of safety, smoothness and continuity, a novel cost function was designed, and the optimal path was selected by minimizing the cost function. Various simulative roads were designed to test the proposed method in the experiments. The experimental results show that the proposed approach has the ability of planning a safe and smooth path for avoiding the obstacles on both single-lane roads and multi-lane roads with good real-time performance.

Endpoint prediction method for steelmaking based on multi-task learning

CHENG Jin, WANG Jian

2017, 37(3):  889-895.  DOI: 10.11772/j.issn.1001-9081.2017.03.889

Asbtract ( )   PDF (1088KB) ( )  

References | Related Articles | Metrics

The quality of the molten steel is usually judged by the hit rate of the endpoint. However, there are many influencing factors in the steelmaking process, and it is difficult to accurately predict the endpoint temperature and carbon content. In view of this, a data-driven Multi-Task Learning (MTL) steelmaking endpoint prediction method was proposed. Firstly, the input and output factors of steelmaking process were analyzed and extracted, and a number of sub-learning tasks were selected to combine the two-stage blowing characteristics of steelmaking. Secondly, according to the relativity between the sub-tasks and the endpoint parameters, the appropriate subtasks were selected to improve the accuracy of the endpoint prediction, and the multi-task learning model was constructed, and the model output was optimized twice. Finally, the process parameters of the multitask learning model were obtained by model training of the processed production data through the proximal gradient algorithm. In the case of a steel plant, compared with neural network, the prediction accuracy of the method proposed increased 10% when endpoint temperature error was less than 12℃ and carbon content error was less than 0.01%. The prediction accuracy increased by 11% and 7% respectively with the error range less than 6℃ and 0.005%. The experimental results show that multi-task learning can improve the accuracy of endpoint prediction in practice.

Multi-pose face reconstruction and recognition based on multi-task learning

OUYANG Ning, MA Yutao, LIN Leping

2017, 37(3):  896-900.  DOI: 10.11772/j.issn.1001-9081.2017.03.896

Asbtract ( )   PDF (881KB) ( )  

References | Related Articles | Metrics

To circumvent the influence of pose variance on face recognition performance and considerable probability of losing the facial local detail information in the process of pose recovery, a multi-pose face reconstruction and recognition method based on multi-task learning was proposed, namely Multi-task Learning Stacked Auto-encoder (MtLSAE). Considering the correlation between pose recovery and retaining local detail information, multi-task learning mechanism was used and sparse auto-encoder with non-negativity constraints was introduced by MtLSAE to learn part features of the face when recovering frontal images using step-wise approach. And then the whole net framework was learned by sharing parameters between above two related tasks. Finally, Fisherface was used for dimensionality reduction and extracting discriminative features of reconstructed positive face image, and the nearest neighbor classifier was used for recognition. The experimental results demonstrate that MtLSAE achieves good pose reconstruction quality and makes facial local texture information clear; on the other hand, it also achieves higher recognition rate than some classical methods such as Local Gabor Binary Pattern(LGBP), View-Based Active Appearance (VAAM) and Stacked Progressive Auto-encoder (SPAE).

Face recognition based on sparse representation and elastic network

LI Guangzao, WANG Shitong

2017, 37(3):  901-905.  DOI: 10.11772/j.issn.1001-9081.2017.03.901

Asbtract ( )   PDF (822KB) ( )  

References | Related Articles | Metrics

Because of the successful use of the sparse representation in face classification algorithm, a more efficient classification method based on Sparse Representation-based pattern Classification (SRC) and elastic network was proposed. To enhance the ability of collaborative representation and enhance the ability to deal with strongly correlated data, a sparse decomposition method based on elastic network was proposed based on the iterative dynamic culling mechanism. Test samples were represented by a linear combination of training samples, and the iterative mechanism was used to remove the categories and samples with less contribution to the classification from all the samples, the Elastic Net algorithm was used for coefficient decomposition to select the samples and classes with high contribution to the classification. Finally, the test samples were classified according to the similarity. The experiment results show that the recognition rate of the algorithm is 98.75%, 86.62% and 99.72% respectively for the ORL, FERET and AR data sets which shows the effectiveness of the proposed algorithm. Compared with the methods of LASSO and SRC-GS, the proposed algorithm can enhance the ability of dealing with high-dimension small sample and strongly correlated variable data in the process of coefficient decomposition. It highlights the importance of sparse constraint in the algorithm and has higher accuracy and stability, and can be more effectively applied to face classification.

Vocal effort in speaker recognition based on MAP+CMLLR

HUANG Wenna, PENG Yaxiong, HE Song

2017, 37(3):  906-910.  DOI: 10.11772/j.issn.1001-9081.2017.03.906

Asbtract ( )   PDF (847KB) ( )  

References | Related Articles | Metrics

To improve the performance of recognition system which is influenced by the change of vocal effort, in the premise of a small amount of whisper and shouted speech data in training speech data, Maximum A Posteriori (MAP) and Constraint Maximum Likelihood Linear Regression (CMLLR) were combined to update the speaker model and transform the speaker characteristics. MAP adaption method was used to update the speaker model of normal speech training, and the CMLLR feature space projection method was used to project and transform the features of whisper and shouted testing speech to improve the mismatch between training speech and testing speech. Experimental results show that the Equal Error Rate (EER) of speaker recognition system was significantly reduced by using the proposed method. Compared with the baseline system, MAP adaptation method, Maximum Likelihood Linear Regression (MLLR) model projection method and CMLLR feature space projection method, the average EER is reduced by 75.3%, 3.5%, 72%, 70.9%, respectively. The experimental results prove that the proposed method weakens the influence on discriminative power for vocal effort and makes the speaker recognition system more robust to vocal effort variability.

Simultaneous range and speed measurement by vehicle laser radar based on pseudo-random noise code modulation

ZHENG Gang, CHENG Yongzhi, MAO Xuesong

2017, 37(3):  911-914.  DOI: 10.11772/j.issn.1001-9081.2017.03.911

Asbtract ( )   PDF (712KB) ( )  

References | Related Articles | Metrics

To solve the problems of high cost and low echo signal utilization efficiency in recently proposed laser radar systems integrated with double optical receivers for measuring range and speed of road targets, a pulsed Doppler laser radar system based on Pseudo-random Noise (PN) code modulation was proposed. The possibility of measuring range and speed simultaneously by the system integrated with single optical receiver was studied. The performance of the proposed method was verified by computer simulation. Firstly, the system model of vehicle laser radar was demonstrated and the existing problem for measuring range and speed by the model was analyzed when it works in pulsed mode. Then, the schematic diagram method for range and speed measurement by analyzing electric signal output from single optical heterodyne receiver was discussed, i.e. the correlation function of the electric signal and local modulation codes was computed for obtaining light flight time, and then target range; the spectrum of the electric signal was computed by non-uniformly sampled signal spectrum analysis method for obtaining Doppler frequency, and then the target speed. Finally, the stability of the proposed method for range and speed measurement was verified by computer simulation. The experimental results show that the method achieves stable range and speed measurement in road environments. Compared to direct detection system, the sensitivity of measurement is improved over 10 dB, which has no relation to echo arrival time and amount of Doppler frequency.