Computer Science

Research on Task Scheduling in Cloud Computing

MA Xiao-jin, RAO Guo-bin, XU Hua-hu

Computer Science. 2019, 46 (3): 1-8. doi:10.11896/j.issn.1002-137X.2019.03.001

Abstract

PDF(1598KB) ( 2155 )

References | Related Articles | Metrics

In cloud computing,virtualization technology separates various kinds of computing resources from the underlying infrastructure and expands them dynamically,and it allows users to pay on the basis of usage.Cloud platform is a heterogeneous system which consists of different hardware and huge data resources.With the increasing number of tasks,it is critical to schedule users’ tasks and allocate resources effectively through task scheduling algorithm.This paper illustrated a brief introduction of cloud computing,task scheduling algorithm and the core scheduling process including evaluation metrics with some figures.Then,it proposed an overview of the related literatures and algorithms in recent years.Finally,this paper presented some key aspects of the research.In realistic applications due to the varying si-tuation of tasks and uncertainty in resources,it is crucial to select the scheduling strategy accordingly,and taking more performance indicators into consideration can enhance the efficiency and quality of service in cloud computing.

Survey of Distributed Machine Learning Platforms and Algorithms

SHU Na,LIU Bo,LIN Wei-wei,LI Peng-fei

Computer Science. 2019, 46 (3): 9-18. doi:10.11896/j.issn.1002-137X.2019.03.002

Abstract

PDF(1744KB) ( 8025 )

References | Related Articles | Metrics

Distributed machine learning deploys many tasks which have large-scale data and computation in multiple machines.For improving the speed of largek-scale calculation and less overhead effectively,its core idea is “divide and conquer”.As one of the most important fields of machine learning,distributed machine learning has been widely concerned by researchers in each field.In view of research significance and practical value of distributed machine learning,this paper gave a summarization of mainstream platforms like Spark,MXNet,Petuum,TensorFlow and PyTorch,and analyzed their characteristics from different sides.Then,this paper made a deep explain for the implementation of machine learning algorithm from data parallel and model parallel,and gave a view of distributed computing model from bulk synchronous parallel model,asynchronous parallel model and delayed asynchronous parallel model.Finally,this paper discussed the future work of distributed machine learning from five aspects:improvement of platform,algorithms optimization,communication of networks,scalability of large-scale data algorithms and fault-tolerance.

Survey on Adaptive Random Testing by Partitioning

LI Zhi-bo, LI Qing-bao, YU Lei, HOU Xue-mei

Computer Science. 2019, 46 (3): 19-29. doi:10.11896/j.issn.1002-137X.2019.03.003

Abstract

PDF(1966KB) ( 1368 )

References | Related Articles | Metrics

As a fundamental software testing technique,random testing (RT) has been widely used in practice.Adaptive random testing (ART),an enhancement of RT,performs better than original RT in terms of fault detection capability.Firstly,this paper analyzed the classical ART algorithm with high detection effectiveness and large time overhead.Se-condly,it summarizedthe ART algorithms by partitioning to reduce the time cost,analyzed and compared various partition strategies and test case generation algorithms.Meanwhile,this paper analyzed the problems of the key factors affecting the effectiveness of ART algorithm and leading to low efficiency of algorithm in high dimensional input domain.Finally,it discussed the problems and challenges in the ART algorithm.

Comprehensive Review of Grey Wolf Optimization Algorithm

ZHANG Xiao-feng, WANG Xiu-ying

Computer Science. 2019, 46 (3): 30-38. doi:10.11896/j.issn.1002-137X.2019.03.004

Abstract

PDF(1416KB) ( 4888 )

References | Related Articles | Metrics

Grey wolf optimization (GWO) algorithm is a new kind of swarm-intelligence-based algorithm and some significant developments have been made since its introduction in 2014.GWO has been successfully applied in a variety of fields due to its simplicity and efficiency.This paper provided a complete survey on GWO,including its search mechanism,implementation process,relative merits,improvements and applications.The studies on GWO about its improvements including improvement of population initialization,search mechanism,and parameters were especially discussed.The application status of GWO in aspect of parameter optimization combinatorial optimization and complex function optimization was summarized.Finally,some novel research directions for future development of this powerful algorithm were given.

Survey on Short-term Traffic Flow Forecasting Based on Deep Learning

DAI Liang,MEI Yang,QIAO Chao,MENG Yun,LV Jin-ming

Computer Science. 2019, 46 (3): 39-47. doi:10.11896/j.issn.1002-137X.2019.03.005

Abstract

PDF(1685KB) ( 2726 )

References | Related Articles | Metrics

Short-term traffic flow forecasting is a hot topic in the field of intelligent transportation,which is of great significance in traffic control and management.The traditional traffic flow forecasting methods are difficult to describe the internal characteristics of the traffic data accurately.Deep learningcan learn the internal complex multivariate coupled structure of the traffic flow data through its deep structure and then make a more accurate forecasting of the traffic flow,which makes deep learning a hot topic in the current traffic flow forecasting field.Firstly,the traditional traffic flow forecasting methods and the current research status of deep learning were briefly introduced.Then the methods of short-term traffic flow forecasting based on deep learningwere classified according to generative deep architecture and discriminative deep architecture.This paper also summarized the main methods of deep learning in the field of traffic flow forecasting and compared their performance.Finally,the existing problems and development directions of deep learning in short-term traffic flow forecasting were discussed.

Review of Bottom-up Salient Object Detection

WU Jia-ying,YANG Sai,DU Jun,LIN Hong-da

Computer Science. 2019, 46 (3): 48-52. doi:10.11896/j.issn.1002-137X.2019.03.006

Abstract

PDF(2305KB) ( 1712 )

References | Related Articles | Metrics

This paper reviewed the current development status at home and abroad in the field of salient object detection.Firstly,this paper introduced the research background and development process of salient object detection.Then,aiming at the difference of the features used by each saliency model,it summarized the saliency calculation from two aspects of hand-crafted features and deep learning features.While the saliency calculation based on hand-crafted features are addressed,it is further classified into the following three subcategories,i.e.the saliency calculation based on contrast prior,the saliency calculation based on foreground prior,and the saliency calculation based on back ground prior.Meanwhile,this paper elaborated the basic ideas of saliency modeling in each subcategory.Finally,it discussed the problems to be solved and further research directions of salient object detection.

Survey on Non-frontal Facial Expression Recognition Methods

JIANG Bin,GAN Yong,ZHANG Huan-long,ZHANG Qiu-wen

Computer Science. 2019, 46 (3): 53-62. doi:10.11896/j.issn.1002-137X.2019.03.007

Abstract

PDF(1375KB) ( 1452 )

References | Related Articles | Metrics

Facial expression recognition is an important part in biometric feature recognition,and it is also a key techno-logy of human-machine interaction.However,most methods only focus on the frontal or nearly frontal facial images and videos,and restrict the normal head movements,so it is bad for intelligent development of facial expression recognition.To handle this problem,firstly,the face detection,head pose estimation,facial expression feature extraction and classification methods were introduced for exploring the development of non-frontal facial expression recognition system.Se-condly,the non-frontal facial expression feature extraction and classification methods were emphatically introduced,and the comparison and analysis of the facial key points-based non-frontal facial expression recognition algorithm,the appea-rance feature-based non-frontal facial expression recognition algorithm and the pose-depend-based non-frontal facial expression recognition algorithm were carried out.Finally,the current research on the non-frontal facial expression recognition was summarized,and the future research and development direction were prosected.

Review on Development of Convolutional Neural Network and Its Application in Computer Vision

CHEN Chao, QI Feng

Computer Science. 2019, 46 (3): 63-73. doi:10.11896/j.issn.1002-137X.2019.03.008

Abstract

PDF(2702KB) ( 4639 )

References | Related Articles | Metrics

In recent years,deep learning has achieved a series of remarkable research results in various fields such as computer vision,speech recognition,natural language processing and medical image processing.In different types of deep neural networks,convolution neural network has obtained most extensive study,not only reflecting the prosperity in aca-demic field,but also making a tremendous realistic impact and commercial value on the related industries.With the rapidgrowth of annotation sample data sets and the drastic improvement of GPU performance,related researches on convolutional neural networks are rapidly developed and have achieved remarkable results in various tasks in the field of computer vision.This paper reviewed the history of convolution neural network firstly.Then it introduced the basic structure of convolutional neural network and the function of each component.Next,it described the improvements of convolution neural network in convolution layer,pooling layer and activation functionin detail.Also,it summarized typical neural network architectures since 1998(such as AlexNet,ZF-Net,VGGNet,GoogLeNet,ResNet,DenseNet,DPN and SENet).In the field of computer vision,this paper emphatically introducedthe latest research progresses of convolution neural network in image classification / localization,target detection,target segmentation,target tracking,behavior re-cognition and image super-resolution reconstruction.Finally,it summarized the problems and challenges to be solvedabout convolutional neural network.

Review of Generative Adversarial Network

CHENG Xian-yi,XIE Lu,ZHU Jian-xin,HU Bin,SHI Quan

Computer Science. 2019, 46 (3): 74-81. doi:10.11896/j.issn.1002-137X.2019.03.009

Abstract

PDF(3854KB) ( 2935 )

References | Related Articles | Metrics

Humans can understand the way of movement,so they can predictthe future development of things more accurately than machines.But GAN (Generative Adversarial Network) is a new neural Network system,its dataare very lifelike,even people can’t identify whether the data are real or generated.In a sense,GAN provides a brand new thought for guiding the artificial intelligence system to accomplish complex tasks,and makes the machine a specialist.In this paper,first of all,the basic model and some improvements model of GAN were discussed.Then,some application achievements of GAN were shown,such as the images generated by the super resolution,by a text description,by the artistic style and short video generated.Finally,some problems of theory,architecture,and application in the future research were discussed

Research on Ship Detection Technology Based on Optical Remote Sensing Image

YIN Ya, HUANG Hai, ZHANG Zhi-xiang

Computer Science. 2019, 46 (3): 82-87. doi:10.11896/j.issn.1002-137X.2019.03.010

Abstract

PDF(1261KB) ( 2477 )

References | Related Articles | Metrics

The detection of ships in optical remote sensing images is a research hotspot with broad applications in the field of remote sensing image processing and analysis.Based on the optical remote sensing image,this paper summarized the main processing methods used in each link around the general processing flow of ship detection,and analyzed the advantages and disadvantages of each method.Then,the paper pointed out the bottleneck problems faced by each link,expounded the limitations of natural image detection methods applied to ship target detection and discussed the challenges faced by current research.Finally,the relevant development trends were discussed.

Deep Residual Network Based HEVC Compressed Videos Enhancement

HE Xiao-yi,DUAN Ling-yu,LIN Wei-yao

Computer Science. 2019, 46 (3): 88-91. doi:10.11896/j.issn.1002-137X.2019.03.011

Abstract

PDF(2331KB) ( 1436 )

References | Related Articles | Metrics

This paper proposed a HEVC-compressed videos enhancement method based on deep residual network.This method utilizes several stacked residual blocks to achieve feature extraction,followed by feature enhancement and reconstruction.Compared with the existing methods which only use a few convolutional layers,the proposed method can capture the feature of input compressed frames in a more distinctive and stable way.Experimental results show that the proposed method leads to over 6.92% BD-rate saving on 20 benchmark sequences and achieves the best performance among the compared methods.

Perceptual Model Based on GLCM Combined with Depth

YE Peng, WANG Yong-fang, XIA Yu-meng, AN Ping

Computer Science. 2019, 46 (3): 92-96. doi:10.11896/j.issn.1002-137X.2019.03.012

Abstract

PDF(1545KB) ( 813 )

References | Related Articles | Metrics

Just Noticeable Distortion (JND) model is a kind of perceptual model,which is one of the most effective methods to remove the visual redundancy in image/video compression.Because the calculation of the contrast masking effect (CM) is not perfect and the consideration of depth information is not accurate in the existing JND model,this paper proposed a JND model combined with depth based on gray level co-occurrence matrix (GLCM).Firstly,the image is decomposed into the edge part and the texture part by the total variance(TV) method,the edge part is processed by Canny operator and the texture part is processed by GLCM.A more accurate CM model is formed by incorporating above two parts.Further,a new JND model based on gray-level co-occurrence Matrix is established by combining the background brightness masking effect.Besides,based on the human depth perception,a novel depth weighting model is proposed.Finally,a new perceptual model combined with depth based on GLCM is established.The experimental results show that the proposed model is more consistent with the human visual perception.Comparing with the existing JND model,the proposed model can tolerate more distortion and has much better perceptual quality.

Improved MDP Tracking Method by Combining 2D and 3D Information

WANG Zheng-ning, ZHOU Yang, LV Xia, ZENG Fan-wei, ZHANG Xiang, ZHANG Feng-jun

Computer Science. 2019, 46 (3): 97-102. doi:10.11896/j.issn.1002-137X.2019.03.013

Abstract

PDF(2271KB) ( 787 )

References | Related Articles | Metrics

Online multi-object tracking (MOT) plays an important role in autonomous driving and ADAS system.Most of recent MOT methods concentrate on tracking in image domain.Although they can solve most of problems by building adaptive online models or optimizing energy functions,it’s still an obstacle for researchers to handle mutual occlusion in complex traffic scenes.In this paper,an improved tracking method was proposed by introducing 3D information to the Markov decision processes (MDP) tracker.The original MDP similarity feature was extended from image domain to spatial domain with 2D-3D combined feature,and a new optical flow descriptor,called multi-image FB error,was addressed to replace the original multi-aspect FB error.This methodwas tested on KITTI benchmark and the results verified that the comprehensive performance of the proposed method is refined significantly in comprehensive performance compared with the original method.

Profit Optimization for Multi-content Video Streaming over Mobile Network Based on User Preference

XU Jing-ce, LIANG Bing, LI Meng-nan, JI Wen, CHEN Yi-qiang

Computer Science. 2019, 46 (3): 103-107. doi:10.11896/j.issn.1002-137X.2019.03.014

Abstract

PDF(1477KB) ( 823 )

References | Related Articles | Metrics

In recent years,the emergence of 4G and 5G network has greatly improved the bandwidth of mobile device data transmission,while the performance of video playback devices has been also improved,which increases the user’s demand on the quality of video streaming gradually.Thus,improving the profit of video streaming over mobile network is becoming more and more important.This paper analyzed the effect of user preference on the profit of multi-content videostreaming system.Moreover,this paper proposed the profit model of End Users based on user preference by consi-dering the traffic cost and formulated the optimization problem of total system profit into weighted profit optimization problem.Considering that the users with different preferences have different effects on the total profit of video streaming system,this paper proposed a weight selection algorithm of End Users based on preference-bitrate ratio to select the optimal weights under the condition of current user preferences.Then the optimal bitrate under the condition of current user preference was obtained by solving the optimization problem of optimal weighted profit of End Users.The experimental results show that the proposed method improves the total profit of system by 5％～10% compared with the exis-ting method.

Adaptive Weighted Bi-prediction Method Based on Reference Quality

YANG Min-jie, ZHU Ce, GUO Hong-wei, JIANG Ni

Computer Science. 2019, 46 (3): 108-112. doi:10.11896/j.issn.1002-137X.2019.03.015

Abstract

PDF(1340KB) ( 794 )

References | Related Articles | Metrics

In modern video codecs,bi-prediction technique plays a significant role for removing temporal redundancy by exploiting temporal correlations between pictures.The bi-prediction signal is formed simply by averaging two uni-prediction signals using a fixed weight value 0.5.However,it will produce serious distortion in some condition that illumination changes rapidly from one reference picture to another or the prediction quality of one motion-compensated prediction block may differ from the other due to the factors such as quantization.To solve the above problems,an adaptive weighted bi-prediction method based on reference quality was proposed in this paper.In this scheme,the greater weight value will be assigned to the reference block if the quality of the reference block is better,and vice versa.The simulation results show that compared with JEM5.0.1,the proposed weighted bi-prediction can achieve about 0.25% and 0.3% Bjntegaard delta (BD) bitrate savings on average under random access main (RA) and low-delay B main(LDB) confi-gurations,respectively,while the increased encoding and decoding complexities are moderate.

Deep Learning Based Fast VideoTranscoding Algorithm

XU Jing-yao, WANG Zu-lin, XU Mai

Computer Science. 2019, 46 (3): 113-118. doi:10.11896/j.issn.1002-137X.2019.03.016

Abstract

PDF(1725KB) ( 1085 )

References | Related Articles | Metrics

Due to the good rate-distortion performance,as the latest video compression standard,high efficiency video coding (HEVC) has been adopted by more and more terminals.However,there are still a large number of H.264 streams in the field of video compression.Therefore,H.264 to HEVC video transcoding is a meaningful research issue.The simplest way to achieve H.264 to HEVC transcoding is to directly cascade the H.264 decoder and the HEVC encoder.Due to high complexity of the HEVC coding process,this transcoding method is time-consuming.Therefore,this paper proposed a fast H.264 to HEVC transcoding method based on deep learning to predict the CTU(Coding Tree Unit) partition of HEVC,avoiding the brute-force search of CTU partition for rate-distortion optimization(RDO).First,a large-scale database of H.264 to HEVC transcoding is built for ensuring the training of deep learning model.Second,the correlation between HEVC CTU partition and H.264 domain features is analyzed,and the similarity of CTU partition across frames is found out.Then,a three-level classifier based on LSTM (Long Short-Term Memory) is designed to predict the CTU partition.The experimental results show that the H.264 to HEVC fast transcoding algorithm proposed in this paper achieves 60% reduction in complexity compared to the original transcoder,while the peak signal-to-noise ratio is only reduced by 0.039kdB,so the proposed method outperforms the state-of-the-art transcoding methods.

Deep Convolutional Prior Guided Robust Image Separation Method and Its Applications

JIANG Zhi-ying, LIU Ri-sheng

Computer Science. 2019, 46 (3): 119-124. doi:10.11896/j.issn.1002-137X.2019.03.017

Abstract

PDF(3044KB) ( 939 )

References | Related Articles | Metrics

Single image layer separation aims to divide the observed image into two independent and practical components based on the requirement of tasks.Many tasks in computer vision can be understood as the separation of two different layers essentially,such as single image rain streak removal,intrinsic image decomposition and reflection removal.Therefore,an excellent image layer decomposition method would promote the solution of these problems greatly.Since there is only one known variable,two variables need to be recovered.This problem is fundamentally ill-posed.Most exis-ting approaches tend to design complex priors according to the different characteristics between the two separated layers.However,loss function with complex prior regularization is hard to be optimized.Performance is also compromised by the fixed iteration schemes and less data fitting ability.More importantly,these conventional prior based methods can only be applied to one specific task as they are weak in generalization.To partially mitigate the limitations mentioned above,this paper developed a flexible optimization technique to incorporate deep architectures into optimization iterations for adaptive image layer separation.As we all know,the convolutional neural network is a network structure composed of convolutions and other non-linear operations.

第3期

姜智颖,等:深度卷积先验引导的鲁棒图像层分离方法及其应用

The convolution operation uses different convolution kernels to extract different features for a given image,so the convolution kernel has very strong capabilities for feature extraction.Recently,the advantages of deep learning in feature extraction have been gradually reflected and are increasingly used in the low-level image processing.Therefore,the proposed method uses deep convolutional prior instead of traditional model prior to characterize different layers.At the same time,in order to reduce the network’s dependence on training data and improve the effectiveness of the algorithm on different tasks,deep information is combined with traditional optimization framework.Specifically,energy function using MAP (Maximum A Posteriori) is built and then the model is transfered to three subproblems based on ADMM (Alternating Direction Method of Multipliers).The first two subproblems are to estimate two approximate separated layers,and the other subproblem is to solve the final result.In other words,deep convolutional networks are used to guide the process of model optimization.In this way,the proposed method not only retains the advantage of feature extraction in deep structure,but also maintains the stability of traditional model optimization and improves the effectiveness of networks.Finally,this method is applied to a variety of ima-ge restoration tasks,including single image rain streak removal and reflection removal.By comparing this method with several tasks-specific methods including conventional model methods and deep learning methods respectively,this me-thod shows great advantages in both visual effects and numerical results.It reveals that this method has a strong genera-lization in multi-tasks and outperforms other methods in each task.

Liver CT Image Feature Extraction Method Based on Improved Multi-scale LBP Algorithm

LIU Xiao-hong, ZHU Yu-quan, LIU Zhe, SONG Yu-qing, ZHU Yan, YUAN De-qi

Computer Science. 2019, 46 (3): 125-130. doi:10.11896/j.issn.1002-137X.2019.03.018

Abstract

PDF(2200KB) ( 1285 )

References | Related Articles | Metrics

Liver cancer,Malignant liver tumors,can be divided into primary and secondary categories.Recent census data prove that the current annual mortality of liver cancer has ranked third in the world.The diagnosis of early liverdi-sease is beneficial to the treatment of liver cancer.The local binary pattern(LBP) algorithm has been widely used in the diagnosis of liver lesions.Although the traditional LBP method is simple,efficient,and easy to understand,but it lacks multi-scale information which leads to incomplete information description and lack of key information.In view of the defect that high order directional derivative local binary pattern(DLBP) algorithm will lose key information,extended multi-scale LBP algorithm(MSLBP) was proposed.The method firstly preprocesses the liver CT image to extract ROI region,then uses the extended multi-scale LBP feature extraction method to extract features.This method fuses the high-order sampling point information with its neighboring point information as the final information of the sampling point to participate in the operation.At the same time,the operation of averaging the diagonal regions highlights the neighborhood and describes the texture information of the liver image from a larger range.Finally,the classification algorithm is executed.The experimental results show that the accuracy of the proposed method can reach 90.1%,which is 8.7% higher than the original LBP feature extraction method.

第3期

刘晓虹,等:基于改进多尺度LBP算法的肝脏CT图像特征提取方法

It has certain clinical application significance and can be used to help doctors diagnose.In the image preprocessing section,since medical images are different from natural images,the DICOM images gotten from hospital cannot be used directly.The first step of image preprocessing is to set Pixel Padding Value to zero.The second step of image preprocessing is converting pixel values to CT values using the equation 7 in section 2.1 according to header file information of the DICOM image.Then,an improved multi-scale LBP feature extraction was performed.The multi-scale feature is extracted while the relationship between neighboring pixels is considered.The LBP model used in this paper is a uniform LBP,with a total of 59 features.In order to prove the effectiveness of the improved multi-scale algorithm,this paper used complete local binary pattern(CLBP),four-patch LBP(FPLBP),dominant rotated local binary pattern(drLBP),local binary pattern(LBP) and other feature extraction methodsto extract the texture features of liver CT images,and then compared the experimental results,as shown in Table 1 in Section 4.2.Through the statistics of feature dimensions for all methods,it is proved that the multi-scale LBP method proposed in this paper has low dimensionality and high efficiency.The experimental results show that the proposed method can extend the multi-scale characteristics of LBP well,and describe the macro-texture structure information of a larger area while maintaining the same dimension.At the same time,the relationship information between adjacent pixels is taken into account,which makes up for the lack of sufficient information description and improves the accuracy of the algorithm.

Video Advertisement Classification Method Based on Shot Segmentation and Spatial Attention Model

TAN Kai, WU Qing-bo, MENG Fan-man, XU Lin-feng

Computer Science. 2019, 46 (3): 131-136. doi:10.11896/j.issn.1002-137X.2019.03.019

Abstract

PDF(3571KB) ( 830 )

References | Related Articles | Metrics

As video advertisement is increasingly used in some areas such as search and user recommendation,advertisement video classification becomes an important issue and poses a significant challenge for computer vision.Different from the existing video classification task,there are two challenges of advertisement video classification.First,advertised products appear in advertisement video aperiodically and sparsely.This means that most of frames are irrelevant to advertisement category,which can potentially cause interference with classification models.Second,there are complex background in advertisement video which makes it hard to extract useful information of product.To solve these problems,this paper proposed an advertisement video classification method based on shot segmentation and spatial attention model (SSSA).To address interference of irrelevant frames,a shot based partitioning method was used to sample frames.To solve the influence of complex background on feature extraction,the attention mechanism was embedded into SSSA to locate products and extract discriminative feature from the attention area which is mostly related to the advertised products.An attention predictionnetwork (APN) was trained to predict the attention map.To verify the proposed model,this paper introduced a new thousand-level dataset for advertisement video classification named TAV,and the gaze data were also collected to train the APN.Experiments evaluated on the TAV dataset demonstrate that the performance of the proposed model improves about 10％ compared with the state-of-the-art video classification methods.

Real-time High-confidence Update Complementary Learner Tracking

FAN Rong-rong, FAN Jia-qing, LIU Qing-shan

Computer Science. 2019, 46 (3): 137-141. doi:10.11896/j.issn.1002-137X.2019.03.020

Abstract

PDF(3056KB) ( 959 )

References | Related Articles | Metrics

To address the issue that the complementary learner tracking algorithm (Staple) cannot perform well when the target suffers from severe occlusions,a high-confidence update complementary learner tracker (HCLT) was proposed.Firstly,at the input frame,a standard correlation filter is employed to calculate the correlation filter (CF) response.Secondly,the confidence value based on the CF response is calculated and the update of the correlation filter is stopped when the current confidence value exceeds the mean confidence value.Then,if the number of the continuous no-updated frames comes up to ten,the tracker will be forced to update the filter.Finally,the final response is obtained by combining the CF response with the color response,and the location of maximum response is the tracking result.Expe-riment results show that compared with several state-of-the-art trackers including complementary learner(Staple),end-to-end representation correlation filter net tracker(CFNet),attentional correlation filter network tracker(ACFN) and hedged deep tracking(HDT),the proposed algorithm is the best in terms of success rate,outperforming the baseline tracker Staple by 1.0 percentage points and 0.4 percentage points interms of success rate and expected average overlap(EAO)on OTB100 dataset and VOT2016 dataset,respectively.Besides,the performance on heavy occlusion and severe illumination variation sequences demonstrates the effectiveness of proposed tracker when handling drastic appearance variations.

Improved R-λ Model Based Rate Control Algorithm

GUO Hong-wei, LUO Hong-jun, LIU Shuai, NIU Lin, YANG Bo

Computer Science. 2019, 46 (3): 142-147. doi:10.11896/j.issn.1002-137X.2019.03.021

Abstract

PDF(1455KB) ( 836 )

References | Related Articles | Metrics

Rate control is an important module in video coding systems,which makes an encoder output specific bit rates,and minimizes the distortion of encoded videos.The R-λ model based rate control is recommended in high efficiency video coding (HEVC) international video coding standard,which mainly includes two schemes,i.e.,the fixed ratio bit allocation and the adaptive ratio bit allocation.In order to improve both the accuracy of rate control and the rate-distortion (R-D) performance,this paper proposed an improved R-λ model based rate control algorithm.Firstly,an accurate R-D model update method is designed according to the coding structure of group of picture (GOP).Secondly,the GOP-level bit allocation scheme is improved according to the relationship of R-D dependency.Finally,the calculation formulas of the dynamic Lagrange multiplier at GOP-level and the dynamic bit weight for the frame to be encoded are proposed.Experimental results demonstrate that the bit rate relative errors of the proposed method are only about 0.006% and 0.005%,and achieves average 1.2% and 1.3% R-D performance gains compared with the adaptive ratio bit allocation scheme under the low delay configuration of P and B frames,respectively.

3-D Model Retrieval Algorithm Based on Residual Network

LI Yin-min, XUE Kai-xin, GAO Zan, XUE Yan-bin, XU Guang-ping, ZHANG Hua

Computer Science. 2019, 46 (3): 148-153. doi:10.11896/j.issn.1002-137X.2019.03.022

Abstract

PDF(2263KB) ( 1070 )

References | Related Articles | Metrics

In recent years,view-based 3D model retrieval has become a key research direction in the field of computer vision.The 3D model retrieval algorithm includes feature extraction and model retrieval where robust features play a decisive rolein retrieval algorithm.Up to now,the traditional hand-crafted features and deep learning features were proposed,but very few people systematically compare them.Therefore,in this work,the performance of different artificial design features and deep learning features was evaluated and analyzed.Based on the premise of full comparison,multiple data sets,multiple evaluation criteria,and different search algorithms were used to conduct experiments.The effects of different layers of deep network on performance were further compared,and a 3D model retrieval algorithm based on residual network was proposed.Several conclusions could be obtained from the experimental results on multiple public datasets.1)When comparing the deep learning features of VGG network and residual network with traditional hand-crafted features,the improvement of comprehensive performance can reaches 3% to 20%.2)Compared with the deep features extracted by VGG network,the comprehensive performance of the residual network is increased by 1% to 5%.3)The performance of different layer features in the VGG network is also different,and the comprehensive performance of the deep and shallow features is increased by 1% to 6%.4)As the depth of the network increase,the overall perfor-mance of the extracted features of the residual network has limited improvement,and is more robust than other contrasting features.

Deinterlacing Algorithm Based on Scene Change and Content Characteristics Detection

ZHU Xiao-tao, LI Yan-ping, HUANG Yuan, HUANG Qian

Computer Science. 2019, 46 (3): 154-158. doi:10.11896/j.issn.1002-137X.2019.03.023

Abstract

PDF(1541KB) ( 700 )

References | Related Articles | Metrics

This paper proposed a deinterlacing algorithm based on scene change and content characteristics detection.Firstly,scene changes and video content characteristics were detected.Secondly,optimized motion estimation was performed based on scene change detection results.Thirdly,the image blocks were locally partitioned and different interpolation methods were applied.Experimental results show that the algorithm can not only improve the vertical image resolution with lower algorithm complexity,but also obtain high-quality progressive sequences for interlaced video sequences of different video content.

Detection Method of Insulator in Aerial Inspection Image Based on Modified R-FCN

ZHAO Zhen-bing, CUI Ya-ping, QI Yin-cheng, DU Li-qun, ZHANG Ke, ZHAI Yong-jie

Computer Science. 2019, 46 (3): 159-163. doi:10.11896/j.issn.1002-137X.2019.03.024

Abstract

PDF(2799KB) ( 920 )

References | Related Articles | Metrics

In the case of partial occlusion of insulator target in aerial inspection images,the region-based fully convolutional networks (R-FCN) model is used for detection,however,the insulator target detection effect is poor and the detection frame cannot completely fit the target.Based on this,this paper proposed an insulator target detection method based on modified R-FCN in aerial inspection image.Firstly,according to the aspect ratio feature of insulator targets,the aspect ratios of proposals in the R-FCN model are modified to 1∶4,1∶2,1∶1,2∶1,4∶1.Then,in view of the occlusion problem in insulator image,an adversarial spatial dropout network (ASDN) layer is introduced into the R-FCN model to generate the samples of incomplete target feature by masking part of feature map,which can improve the detection performance of the model for samples with poor target feature.The average detection rate of R-FCN model reaches 77.27% on the dataset containing 7433 insulator targets.The average detection rate of the modified R-FCN detection method is 84.29%,which improves 7.02%,and the detection frame is more suitable for the target.

Novel Sanitization Approach for Indirect Dependencies in Provenance Graph

SUN Lian-shan, OUYANG Xiao-tong, XU Yan-yan, WANG Yi-xing

Computer Science. 2019, 46 (3): 164-169. doi:10.11896/j.issn.1002-137X.2019.03.025

Abstract

PDF(1400KB) ( 668 )

References | Related Articles | Metrics

Provenance sanitization is a new technology that aims at producing secure provenance views by hiding or redacting sensitive nodes,edges or even indirect dependencies in a provenance graph.However,existing research works mostly focus on sanitizing nodes,rarely on sanitizing edges,not on sanitizing indirect dependencies.To this end,this paper first exemplified the motivations and analyzed the challenges of sanitizing indirect dependencies while keeping utility of provenance views,and formally defined goals and constraints of sanitizing indirect dependencies.Second,this paper proposed a novel mechanism for sanitizing indirect dependencies on the basis of the “Delete＋Repair” mechanism for direct dependency in literature.The proposed mechanism includes both deletion rules and repairing rules.Deletion rules specify what edges can be deleted for breaking all connected paths among two end nodes of a sensitive indirect depen-dency while minimizing the sanitization cost.Repairing rules specify what uncertain dependencies can be added for improving the utility of the sanitized provenance views harmed by applying deletion rules.Finally,a comprehensive sanitization algorithm for sanitizing indirect dependency was implemented and experiments was conducted upon an online open dataset.The experiments results show that the proposed approach can effectively sanitize indirect dependencies while preserving utility of the sanitized provenance view.

Study on Formal Verification of Secure Virtual Machine Monitor

CHEN Hao, LUO Lei, LI Yun, CHEN Li-rong

Computer Science. 2019, 46 (3): 170-179. doi:10.11896/j.issn.1002-137X.2019.03.026

Abstract

PDF(2457KB) ( 1002 )

References | Related Articles | Metrics

Virtualization equips the security-critical systems with multiple features,including partitioning and separation.As the core component,virtual machine monitor (VMM) serves as a backbone to the secure execution as well as a barrier to isolate the threats and faults of virtual machines.Following the principle of least privilege,this paper presented a method to decouple the VMM into two parts:kernel extension and user processes.Furthermore,a formal method by constructing abstraction layers is used to certify those key components of the VMM kernel extension.Then,the functional correctness property of the VMM are also proved.The experiment results show that the certified prototype achieves comparable efficiency as the mainstream virtualization solution.The decoupled design and formal verification improve the VMM security without imposing obvious performance degradation,and meet the requirement of the application fields.

Anomaly Detection Method of Mobile Terminal User Based on Location Information

LI Zhi, MA Chun-lai, MA Tao, SHAN Hong

Computer Science. 2019, 46 (3): 180-187. doi:10.11896/j.issn.1002-137X.2019.03.027

Abstract

PDF(1900KB) ( 1051 )

References | Related Articles | Metrics

Aiming at the problem of trajectory evolution and single-type of detection result in trajectory anomaly detection technology,an anomaly detection method was proposed for mobile terminal user based on location information,which comprehensively utilizes the user historical behavior pattern,group structure information,and behavior of close users.The method converts the location data into the spatio-temporal co-occurrence area(STCOA),and further excavates the user behavior pattern and extractes the user group structure information.On this basis,a multi-class anomaly detection model was constructed by random forest method according to five abnormal characteristics of historical beha-vior pattern anomaly,accompanying behavior pattern anomaly,STCOA behavior pattern anomaly,STCOA flow pattern anomaly and group attribute of abnormal users.This model can identify individual anomaly,group anomaly,spatio-temporal anomaly and event anomaly of mobile terminal users.Experiments on real data sets show that the proposed me-thod can effectively identify the trajectory evolution behavior and detect various types of anomalies of mobile terminal users.Compared with the similar methods,this method has higher recall rate and lower error rate.

Cloud Big Data Integrity Verification Scheme Based on Multi-branch Tree

XIE Si-jiang,JIA Bei,WANG He,XU Shi-cong

Computer Science. 2019, 46 (3): 188-196. doi:10.11896/j.issn.1002-137X.2019.03.028

Abstract

PDF(2079KB) ( 758 )

References | Related Articles | Metrics

With the popularization of the Internet and mobile electronic devices,network storage will become the main way of storage in the future,and the cloud storage also will be the inevitable trend of network storage.How to ensure the integrity of users’ data on cloud storage environment becomes a major problem people concern.Aiming at the problem,this paper presented a cloud big data integrity verification scheme based on multi-branch tree.It realizes public verification with third party auditor and supports privacy-preserving by adding random masking,as well as,it uses a dyna-mic data structure which is multi-branch tree to accomplish dynamic operations.This paper also proposed a new algorithm to get information of data integrity verification from multi-branch tree.Test results show that the scheme can be efficiently applied in the cloud environment to verify data integrity with frequent update operations and multi-users.

Intrusion Detection Based on Semi-supervised Learning with Deep Generative Models

CAO Wei-dong, XU Zhi-xiang, WANG Jing

Computer Science. 2019, 46 (3): 197-201. doi:10.11896/j.issn.1002-137X.2019.03.029

Abstract

PDF(1433KB) ( 964 )

References | Related Articles | Metrics

Aiming at the difficulties that training samples of intrusion detection algorithms based on supervised learning are insufficient,and unsupervised algorithms have low detection rate,a new semi-supervised intrusion detection method based on deep generative models was proposed.This method aims to improve the detection accuracy and the generalization ability of the model by constructing an effective objective function.First,variational auto-encoder in the model is employedto map the vector of raw data from the high-dimensional space to low-dimensional,and the corresponding optimal low-dimension representation of raw can be obtained.Then,the generative model is used to improve the classification accuracy by only using the labeled samples.Experiments show that this method can achieve high accuracy while using a limited number of labeled samples.

Double-auction-based Incentive Mechanism for k-anonymity

TONG Hai,BAI Guang-wei,SHEN Hang

Computer Science. 2019, 46 (3): 202-208. doi:10.11896/j.issn.1002-137X.2019.03.030

Abstract

PDF(1605KB) ( 810 )

References | Related Articles | Metrics

kk-anonymity has become one of the most important location privacy technologies in LBS (Location Based Service).At least k users should be required to build an anonymous set,in which any user cannot be distinguished from other k－1 users.However,many users are not interested in their location privacy,so they have little interest in participating in the construction of anonymous sets.In order to improve the enthusiasm of users to participate in building anonymous sets,this paper proposed a double-auction-based incentive(DAI) mechanism for k-anonymity,which maximizes both the utility of buyers and sellers while guaranteeing fair transaction.To this end,multi-stage sample is used to filter the candidate user sets,then a reasonable remuneration and the winning set of users are determined according to budget balance.Finally,the rationality of the mechanism is provided in consideration of individual rationality,computation efficiency,budget balance and truthfulness,and so on.Simulation results demonstrate that DAI can solve the problem of malicious competition in the existing methods,and improve satisfaction and utility of buyers effectively.

RLWE-based Fully Homomorphic Encryption Scheme with Batch Technique

LI Meng-tian, HU Bin

Computer Science. 2019, 46 (3): 209-216. doi:10.11896/j.issn.1002-137X.2019.03.031

Abstract

PDF(1307KB) ( 1985 )

References | Related Articles | Metrics

The development of information technology and network communication promotes the emergence of big data and cloud computing.The security and privacy of user’s data have gradually become the focus of academic research.Fully Homomorphic Encryption (FHE) is a new research subject,and it has a broad application prospect and important research significance in recent years.It supports arbitrary computation on encrypted data which is equivalent to do the same operations on corresponding plaintext.This feature has important applications in the security of cloud computing.In 2011,Lauter et al. proposed a RLWE-based Homomorphic Encryption scheme,aiming at the scheme,this paper designed a new scheme combined with the batch technique.Concretely,the technique packs multiple “plaintext slots” into a ciphertext by using the Chinese Remainder Theorem,and then performs homomorphic operations on it.Considering the exponential growth of the noise in each multiplication operation,this paper used the key switch and module switch technique to reduce the noise size in ciphertext,which ensure the correct decryption and the next homomorphic computation.Finally,this paper analyzed the security and efficiency of the scheme.It is proved that the proposed scheme is CPA security and the efficiency of encryption is n times to the original scheme.

Users’ Sensitive Information Hiding Method in Hierarchical Heterogeneous Network Based on Mobile Switching Authentication

ZHANG Jian-an

Computer Science. 2019, 46 (3): 217-220. doi:10.11896/j.issn.1002-137X.2019.03.032

Abstract

PDF(1629KB) ( 625 )

References | Related Articles | Metrics

In order to ensure the security of users’ sensitive information,it’s essential to transmit the users’ sensitive information in a covert manner.When the current method is used to hide the users’ sensitive information in the hierarchical heterogeneous network,the obtained information has low integrity and low information security.This paper proposed a users’ sensitive information hiding method in hierarchical heterogeneous networks based on mobile swit-ching authentication.It uses the embedded unit scrambling algorithm to scramble the sensitive information of users and disrupts the normal ordering of users’ sensitive information in hierarchical heterogeneous networks.Then,combined with the HSI model,the optimal users’ sensitive information embedding position is obtained within the scrambled users’ sensitive information sorting sequence according to the character brightness encoding mechanism.And the vector-based manipulation method is used to set the embedding unit of users’ sensitive information in the hierarchical heterogeneous network.According to the analysis results,users’ sensitive information is embedded in the optimal embedding position,and the users’ sensitive information hiding in the hierarchical heterogeneous network is completed.The experimental results show that the information obtained by the proposed method has high integrity and security.

Clustering Assist Feature Alignment for Unsupervised Domain Adaptation

YUAN Ding, WANG Qian, DENG Li-wei

Computer Science. 2019, 46 (3): 221-226. doi:10.11896/j.issn.1002-137X.2019.03.033

Abstract

PDF(1908KB) ( 1240 )

References | Related Articles | Metrics

Supervised deep learning can reach good results in the areas with large amounts of labeled data,but the rea-lity is that there are only a lot of unlabeled data in many areas.How to take advantages of large amounts of unlabeled data has become a key issue in the development of deep learning.Domain adaptation is an effective way to solve this problem.At present,domain adaptation methods based on adversarial training have achieved a good effect.This method uses domain classification loss to align the feature distribution of source domain,and target domain and reduce the difference of distribution between the feature representations of two domain,so the classifier trained with source domain data can be applied to target domain data.The existing domain adaptation method trains the model on the features after domain adaptation and does not make full use of the original information of the target domain data.When the differences between two domains are large,the intra-domain discriminability of target domain features will be reduced.In view of the disadvantages of the present methods,this paper proposed a method for clustering target domain data to assist feature alignment(CAFA-DA) based on the adversarial discriminative domain adaptation (ADDA).Pseudo-labels of target domain data are obtained by clustering and the feature encoder training is constrained in the domain adaptation stage,and the original information of the target domain data is used to improve the discriminability of target domain features.Classifiers trained in the two processes of clustering and domain adaptation are used for ensemble learning and high confidence samples are trained to improve the final effect of the model.The CAFA-DA can be applied to any domain adaption method based on adversarial loss.Finally,this paper compared CAFA-DA with several advanced domain adaption methods on four standard domain adaption data sets.The results show that the accuracy of the CAFA-DA method is better than other methods.Compared with the ADDA method,the results of two comparative experiments are improved by 3.2% and 17.2% respectively.

Cross-modal Retrieval Fusing Multilayer Semantics

FENG Yao-gong CAI Guo-yong

Computer Science. 2019, 46 (3): 227-233. doi:10.11896/j.issn.1002-137X.2019.03.034

Abstract

PDF(2295KB) ( 1218 )

References | Related Articles | Metrics

How to explore the inherent relations of different modalities is the core problem of cross-modal retrieval.The previous works demonstrate that the models which incorporate representation learning and correlation learning into a single process are more suitable for cross-modal retrieval task,but these models only contain the 1-1 correspondence correlations between different modalities.However,different modalities are more likely to have different granularities of semantics abstraction,and the correlations between different modalities are more likely to occur in different layers of semantic at the same time.This paper proposed a cross-modal retrieval model fusing multilayer semantic.The model benefits from the architecture of deep boltzmann machine which is an undirected graph model and implements that each semantic layer of text modality is associated with multiple different semantic layers of image modality at last,and explores the inherent N-M relations of different modalities more sufficiently.The results of experiments on three real and public datasets demonstrate that this model is obviously superior to the state-of-art models,and has higher accuracy of retrieval.

English Automated Essay Scoring Methods Based on Discourse Structure

ZHOU Ming,JIA Yan-ming,ZHOU Cai-lan,XU Ning

Computer Science. 2019, 46 (3): 234-241. doi:10.11896/j.issn.1002-137X.2019.03.035

Abstract

PDF(1553KB) ( 2482 )

References | Related Articles | Metrics

Automated essay scoring is defined as the computer technology that evaluates and scores the composition,based on the technologies of statistics,natural language processing,linguistics and some other fields.Discourse structure analysis is not only an important research field of natural language processing,but also an important component of the AES system.Nowadays,AES system has widely application.However,there is not enough research on the structure of the essay,and the AES system does not focus on the Chinese students.The domestic researches on the AES are in infancy,ignoring the importance of discourse structure in essay scoring.In view of these problems,this paper proposed a method of automated essay scoring based on discourse structure.Firstly,the method extracts essay’s features,such as vocabulary,lexical and discourse structure from levels of words,sentences and paragraphs.Then,the composition of essays is classified by support vector machines,random forests and extreme gradient boosting,and then the linear regression model with the discourse element is constructed to score the compositions.The experimental results show that the accuracy of discourse element identification based random forest (DEI-RF) can reach 94.13%,and the mean squared error of automated discourse structure scoring based on linear regression (DSS-LR) model can reach 0.02,0.11 and 0.08 on introduction,argumentation and concession respectively.

End-to-End Single-channel Automatic Staging Model for Sleep EEG Signal

JIN Huan-huan,YIN Hai-bo,HE Ling-na

Computer Science. 2019, 46 (3): 242-247. doi:10.11896/j.issn.1002-137X.2019.03.036

Abstract

PDF(1953KB) ( 1585 )

References | Related Articles | Metrics

The classification accuracy of current automatic sleep staging is determined by the small data set of imba-lanced classes and hand-engineered features.Aiming at this problem,this paper proposed an automatic sleep staging model based on deep hybrid neural network.For the construction of model’s main structure,the multi-scale Convolutional Neural Networks are used to automatically learn the high-level time-invariant features,the Recurrent Neural Networks constructed by bidirectional Gated Recurrent Unit are used to decode the temporal information from the time invariant features,and the residual connection is used to fully combine the time invariant features with the time information features.For model optimization,in order to reduce the impact of the dataset of imbalanced class on the classification effect of minority class,the experimental data set reconstructed by MSMOTE (Modified Synthetic Minority Oversampling Technique) is used for pre-training.The Swish activation function is used to accelerate the training convergence rate.The experiment was set up on the initial single-channel EEG signal of Fpz-Cz in Sleep-EDF Database.The 15-fold cross-validation experiments show that the overall classification accuracy is 86.85% and the Macro-averaged F1-score is 81.63%.This model can effectively avoid the subjectivity of feature selection and the limitation of class imba-lanced small dataset of imbalanced class in deep learning.

Attribute Reduction for Decision Formal Contexts Based on Threek-way Decision Rules

LIN Hong, QIN Ke-yun

Computer Science. 2019, 46 (3): 248-252. doi:10.11896/j.issn.1002-137X.2019.03.037

Abstract

PDF(1236KB) ( 747 )

References | Related Articles | Metrics

This paper studied the attribute reduction for decision formal context based on three way decision rules induced from three way concept lattice.It proposed the notions of three way decision rules and necessary three way decision rules and discussed their basic properties,presented the concept of consistent set with respect to three way decision rule and examined the judgment theorem for consistent set.Accordingly,based on discernibility matrix and discernibility function,this paper presented the reduction method and an illustrative example.

Community Features Based Balanced Modularity Maximization Social Link Prediction Model

WU Jie-hua,SHEN Jing,ZHOU Bei

Computer Science. 2019, 46 (3): 253-259. doi:10.11896/j.issn.1002-137X.2019.03.038

Abstract

PDF(2107KB) ( 710 )

References | Related Articles | Metrics

Link prediction and community detection are the two major research directions in the field of social network analysis.It is of great significance to explore the community structure and improve the link prediction effect.Based on the modularity maximization link prediction model,this paper proposed a link prediction method based on community structure feature extraction and selection.Firstly,the community structure based similarity index and influence node identification method are introduced into the network evolution model to obtain and link the local and global features respectively.Then,the feature selection algorithm with minimum redundancy and maximum correlation is used to measure the mutual influence,andthe most expressive candidate features are filtered out.Finally,based on the above steps,the features are incorporated into the modularity maximization link prediction model.The algorithm was compared with related algorithms on both artificial and real datasets.The results verify the high efficiency of the algorithm and the necessity of feature extraction and selection based on community structure.

Outlier Detection Algorithm Based on Spectral Embedding and Local Density

LI Chang-jing,ZHAO Shu-liang,CHI Yun-xian

Computer Science. 2019, 46 (3): 260-266. doi:10.11896/j.issn.1002-137X.2019.03.039

Abstract

PDF(2692KB) ( 733 )

References | Related Articles | Metrics

Outlier detection is one of the hot topics in the field of data mining.The existing detection algorithms are mainly applied to the cases where outliers lie in initial attribute subspace or various linear combinations of underlying subspace,when the outliers are embedded in local nonlinear subspace,it is very difficult to detect the outliers effectively.To solve this problem,the shortcomings of typical spectral embedding algorithm for outlier detection were firstly analyzed,and then on the basis of local density,an outlier detection algorithm based on spectral embedding and local density was proposed.The algorithm which uses iterative strategy can efficiently screen unimportant eigenvectors and discover eigenvectors that are relevant for finding outliers hidden in local non-linear subspaces,and the local density-based spectral embedding from a previous iteration is used for improving the similarity graph for the next iteration,such that outliers are gradually segregated from inliers during these iterations.The simulation results show that the detection accuracy of the proposed algorithm is better than other typical algorithms,and it is not sensitive to the parameter setting.

Dam Defect Recognition and Classification Based on Feature Combination and CNN

MAO Ying-chi,WANG Jing,CHEN Xiao-li,XU Shu-fang,CHEN Hao

Computer Science. 2019, 46 (3): 267-276. doi:10.11896/j.issn.1002-137X.2019.03.040

Abstract

PDF(2572KB) ( 1110 )

References | Related Articles | Metrics

Dam defect recognition and classification technology is the basic manifestation of human intelligence.It is one of the most typical and difficult pattern recognition problems.Due to the low signal-to-noise ratio and extremely uneven illumination distribution of dam defects,the recognition rate of classification and recognition algorithm is relatively low.In order to solve these problems,this paper proposed a defect image recognition method based on the combination of ima-ge LBP features and image Gabor features combined with CNN(LBP and Gabor feature combination and CNN,LG-CNN),analyzed the collected dam image,and realized the recognition and classification of the defective images.Firstly,the LBP features and the Gabor features of images are extracted respectively.Then,the features of LBP and Gabor are combined to be the input of CNN.Finally,by training the network layer by layer,the classification and recognition of dam defects are realized.The experimental results show that the average recognition accuracy of LGk-CNN is 88.39%,as well as the recall rate of defect is 92.75%.Compared with the CNN classification algorithm under the same parameters,the recognition accuracy and therecall rate of defect are increased by 3.1% and 2.5% respectively,and the results is the best results.

Frontier Scientific Keyword Extraction Based on Bibliometric and Crowdsourcing

LV Jia-gao,LIANG Kui-yang,CAI Wei

Computer Science. 2019, 46 (3): 275-282. doi:10.11896/j.issn.1002-137X.2019.03.041

Abstract

PDF(1632KB) ( 1700 )

References | Related Articles | Metrics

With the rapid development of science,the annual amount of scientific papers is growing,and new challenge is to extract the frontier scientific keywords from lots of papers.In traditional way,the extraction work is done by experts,which is inefficient and costs much.A new algorithm based on bibliometric analysis and crowdsourcing technique was proposed in this paper.Part-of-speech tagging is used to obtain the nouns from scientific papers,and potentialscie-ntific keywords are selected from these nouns by bibliometric analysis.The last procedure is using data from crowdsourcing platform to check potential scientific keywords and get results.English scientific papers in computer scie-nce and biomedicine are used to conduct experiments.The experiment results suggest that the proposed algorithm has effect on extraction,and it’s more efficient than expert extraction procedure,so it can assist the expert to analysis frontier scientific keywords.In conclusion,this algorithm can do automatic extraction and show possibility of more automatic and intelligent extraction procedure in the future.

Method of Fast Neural Style Transfer with Spatial Constraint

LIU Hong-lin, SHUAI Ren-jun

Computer Science. 2019, 46 (3): 283-286. doi:10.11896/j.issn.1002-137X.2019.03.042

Abstract

PDF(2454KB) ( 851 )

References | Related Articles | Metrics

Gram matrix,a method to get the inner product in simple terms,was commonly used for image style extraction in the style-transfer techniques.The Gram matrix can only extract the static features,but it is completely unconstrained to the spatial sequence of objects in the picture.This paper proposed a fast neural style transfer method with space constraints.First,the residuals are used to redesign the transform network of fast neural style transfer.Then,the method of spatial offset is used to transform the Feature map.Feature map T(a^l) are used for Gram matrix computation to get the cross-correlation,which contains the spatial information.That is to say,it can constrain the object’s spatial sequence in the picture.Finally,experiments show that the method’s ability of space constraint is better than traditional method,and the stylized image with better effect can be quickly obtained.

Heterogeneous Emotional Contagion Model for Crowd Evacuation Simulation

WANG Meng-si, ZHANG Gui-juan, LIU Hong

Computer Science. 2019, 46 (3): 287-297. doi:10.11896/j.issn.1002-137X.2019.03.043

Abstract

PDF(8592KB) ( 1174 )

References | Related Articles | Metrics

In the process of crowd evacuation,individuals tend to aggregate into small groups,while there are also individual differences within any group.However,the existing emotional contagion models do not consider the impact of these factors.In order to improve the authenticity of evacuation simulation,this paper proposed an approach of crowd evacuation based on heterogeneous emotional contagion model by using social comparison theory.First,a computational method of heterogeneous emotional contagion based on Durupinar model was put forward.In the method,an emotional contagion calculating method based on grouping was proposed,which divides the crowd into groups according to the distance and the relationship between individuals,and the impact of grouping on emotional contagion calculation was stu-died.Based on this,a method of calculating the degree of emotional panic based on characteristics was presented,where individuals are specialized according to the characteristics of personality,gender and age,and the degree of emotional panic was calculated by using these characteristics.Then,the proposed method was applied to the process of crowd motion,and heterogeneous emotions were used to drive crowd motion.Finally,the realistic rendering method was deployed to simulate the effect of crowd evacuation.The method of simulating crowd evacuation based on heterogeneous emotio-nal contagion model can effectively simulate the crowd movement in various scenarios.Compared with the previous me-thods,this method considers the influence of grouping and individual characteristics on emotion,and reflects the heterogeneity among individuals.The experimental results show that the method can simulate the crowd behavior more realistically.

Classification of Small Difference Behavior Characteristics Based on Intelligent Vision

CHEN Wei, LIU Yan, LEI Qing

Computer Science. 2019, 46 (3): 298-302. doi:10.11896/j.issn.1002-137X.2019.03.044

Abstract

PDF(3661KB) ( 673 )

References | Related Articles | Metrics

In view of the shortcomings of traditional difference behavior feature classification methods,such as ineffective recognition of small difference behavior and low classification accuracy,a classification method of small difference behavior feature based on intelligent vision was put forward.Firstly,with the immune multi-agent method,the features of small difference behavior are extracted to conduct the immune multi-agent operation for the acquired image set and to analyze the small difference behavior of the slight deformation of the characters to obtain the feature extraction set.Then,the method of video frame image array detection is used to pre-process the gray level of the pixels in the feature extraction set.By constructing the video frame image array,the gray level pixel value is obtained by tracking and recognition initialization learning,and the better small difference behavior feature set is obtained.Finally,the multi-criteria small difference behavior feature classification method is used to segment the better feature set,and the best small difference behavior feature classification results are obtained by contrasting each feature subset with the measurement criteria.The experimental results show that the proposed method improves the classification accuracy of small difference behavior features with a high efficiency.

Multitask Hierarchical Image Retrieval Technology Based on Faster RCNNH

HE Xia, TANG Yi-ping, WANG Li-ran, CHEN Peng, YUAN Gong-ping

Computer Science. 2019, 46 (3): 303-313. doi:10.11896/j.issn.1002-137X.2019.03.045

Abstract

PDF(7919KB) ( 933 )

References | Related Articles | Metrics

Aiming at the problems of low-level automation and intelligence,lack of deep learning,being difficult to obtain high retrieval accuracy,large storage space,slow retrieval speed and hardly meeting the search requirements of big data era for the existing search technologies,this paper proposed a multitask hierarchical image retrieval technology based on faster RCNNH(Faster RCNN Hash).Firstly,the logical regression is performed on the feature map by using the selective retrieval network to obtain the probability vectors of each region of interest in the image.On this basis,the compact quantization network is combined to encode the probability vector and obtain the compact and quantitative hash of the image.Secondly,the re-screening network is utilized to obtain the region-aware semantic features of each region of interest.Then,a precise search strategy based on quantitative hashing matrix is applied into each region of interest to compare the images fast.Finally,the image that is most similar to the corresponding region of interest in the query ima-ge is selected.Meanwhile,the proposed multitask learning method not only can simultaneously obtain compact and quantized hash codes and region-aware semantic features,but also can effectively remove the interference of the background and other objects.The experimental results show that the proposed method can achieve end-to-end training,and the network can automatically select the features with higher quality of the region of interest,thereby improving the automation and intelligence of large-scale image retrieval. The retrieval accuracy (0.9478) and search speed (0.306ks) of the proposed method are both significantly better than the existing large-scale image search technologies.

Estimating Graphlets via Two Common Substructures Aware Sampling in Social Networks

ZHAO Qian-qian, LV Min, XU Yin-long

Computer Science. 2019, 46 (3): 314-320. doi:10.11896/j.issn.1002-137X.2019.03.046

Abstract

PDF(1888KB) ( 813 )

References | Related Articles | Metrics

Graphlets refer to the connected induced subgraphs with small amount of nodes in large-scale network,and have extensive applications in social networks and bioinformatics.Due to extremely high computational costs of exactly counting graphlets,approximately estimating graphlets concentrations via random walk sampling algorithms already becomes the mainstream approach.As node size k increases,the number of k-graphlets increases rapidly and their structures change dramatically,so it is a challenge to quickly estimate the relative frequency of all types of graphlets (graphlet concentrations) in a large-scale network.Aiming at this problem,this paper proposed a novel sampling algorithm,namely common substructures path and 3-star based graphlets sampling via random walk (CSRW2),to efficiently estimate graphlets concentrations.Given k (k=4,5),apart from sampling path via random walk,CSRW2 also samplesano-ther substructure 3-star,and then derives graphlets concentrations by proportional amplification to find the dense graphlets with less appearance more efficiently and adapt to the complex structural changes.Experimental evaluations on real networks demonstrate that CSRW2 can estimate k-graphlets in a uniform framework.CSRW2 outperforms the representative methods in terms of accuracy and is more accurate for the k-graphlets with more edges and less appearances in graphs.For example,when 5-graphlets in sofb-Penn94 is estimated,the average NRMSE of all 5-graphlets is decreased to 0.22 via CSRW2 in contrast to 0.8 obtained by WRW.

Hierarchical Performance Diagnosis Method for Cloud Operating System

YUAN Yue

Computer Science. 2019, 46 (3): 321-326. doi:10.11896/j.issn.1002-137X.2019.03.047

Abstract

PDF(1557KB) ( 674 )

References | Related Articles | Metrics

Recently,quite some researchers aim to develop automatic performance diagnostic tools for dealing with the large-scale and high-load distributed environment.Cloud operating system is the middle layer between cloud user and cloud resource,and diagnosing and settling the problem of slow response of cloud operating system is helpful for optimizing the performance of cloud computing system.It is a challenging job to analyze the performance of executing task in large-scale and complex distributed cloud computing environment.In light of this,this paper proposed a log-based performance diagnosis method for cloud operating system to find out the reason for low execution speed of appointed tasks and provide clues for performance optimization.This method combines the implementation principal of cloud operating system,separates and extracts relevant logs of each executing tasks from the massive logs generated by cloud operating system,and extracts key information,so as to construct hierarchical performance description model and refine the analysis granularity to function executed granularity layer by layer.Finally,through using this method,the main factor of low execution speed can be gotten,which can assist to locate the source of abnormal performance,and it doesn’t need to modify the source code and use the source code to conduct analysis.This paper utilized the OpenStack as prototype system,created the cloud computing environment,and conducted large-scale concurrent simulation experiment.The experimental results demonstrate that the proposed method can provide efficient clues for optimizing system performance and improve the performance obviously,e.g. the consumed time of cloud resource scheduling can be reduced from minute level to second level.

Prefetching Algorithm of Sarsa Learning Based on Space Optimization

LIANG Yuan, YUAN Jing-ling, CHEN Min-cheng

Computer Science. 2019, 46 (3): 327-331. doi:10.11896/j.issn.1002-137X.2019.03.048

Abstract

PDF(1434KB) ( 1053 )

References | Related Articles | Metrics

As the cluster centers for high-performance computers,data centers are busy with CPU clusters.Irregular data structures and algorithms are frequently used,so that most prefetching technologies based on spatio-temporal locality are no longer applicable.This paper referred to the concept of semantic locality,used the reinforcement learning Sarsa algorithm to approximate semantic locations,and predicted irregular data structures and future memory accesses of algorithms.Due to the large state space and action space,this paper used Deep Q-learning method to optimize the State-action space to fit the new state with the old one and took a similar approach if the two states are similar to improve the generalization ability.The experiment on the standard data set SPECCPU 2006 proves that this method has a widegene-ralization ability and can improve the Cache hit rate effectively.

Distributed Online Conditional Gradient Optimization Algorithm

LI De-quan, DONG Qiao, ZHOU Yue-jin

Computer Science. 2019, 46 (3): 332-337. doi:10.11896/j.issn.1002-137X.2019.03.049

Abstract

PDF(1606KB) ( 1147 )