Table of Content

    10 April 2020, Volume 40 Issue 4
    Blockchain technology
    Blockchain-based access control framework for Internet of things
    SHI Jinshan, LI Ru, SONG Tingting
    2020, 40(4):  931-941.  DOI: 10.11772/j.issn.1001-9081.2019111931
    Asbtract ( )   PDF (1700KB) ( )  
    References | Related Articles | Metrics
    The characteristics of massiveness,dynamics,and lightweight devices for network devices in the Internet of Things(IoT)are inherently connected and exist simultaneously. To satisfy these three characteristics simultaneously,a Blockchain-Based IoT Access Control(BBIAC)framework was proposed. Firstly,the BBIAC model under this framework was proposed,the concept of attribute was introduced into the process of IoT authorization to realize the model's support for massiveness;the distributed structure and identity authentication method of blockchain provide the support of dynamics for the model. Secondly,the complete workflow of BBIAC model was introduced. Thirdly,the formal safety assessment of BBIAC was performed by Colored Petri Network(CPN),and the security of the BBIAC model was proved. Experimental results show that BBIAC is suitable for IoT environments with characteristics of massiveness,dynamics and lightweight devices.
    Blockchain enhanced lightweight node model
    ZHAO Yulong, NIU Baoning, LI Peng, FAN Xing
    2020, 40(4):  942-946.  DOI: 10.11772/j.issn.1001-9081.2019111917
    Asbtract ( )   PDF (632KB) ( )  
    References | Related Articles | Metrics
    The inherent chain structure of blockchain means that its data volume grows linearly and endlessly. Over time,it causes a lot of pressure on the storage of the single node,which greatly wastes the storage space of the whole system. The Simplified Payment Verification(SPV)node model proposed in the Bitcoin white paper greatly reduces the node's need for storage space. However,it reduces the number of nodes and increases the pressure,which weakens the decentralization of the entire system and has security risks such as denial of service attacks and witch attacks. By analyzing the Bitcoin block data,a fully functional enhanced lightweight node model Enhanced SPV(ESPV)was proposed. The block was divided into new blocks and old blocks by ESPV,and different storage management strategies were adopted for them. The new block was saved in full copy(one copy per node)for transaction verification,allowing ESPV to has transaction verification(mining) function with less storage space cost. The old block was stored in the nodes of the network in slices,and was accessed through the hierarchical block partition routing table,thereby reducing the waste of the storage space of the system under the premise of ensuring data availability and reliability. The ESPV nodes have full node functionality,thus ensuring the decentralization of the blockchain system and enhancing the security and stability of the system. The experimental results show that the ESPV nodes have more than 80% transaction verification rate,and the data volume and growth amount of these nodes are only 10% of those of all nodes. The data availability and reliability of ESPV are guaranteed,and it is applicable to the whole life cycle of the system.
    Smart contract vulnerability detection scheme based on symbol execution
    ZHAO Wei, ZHANG Wenyin, WANG Jiuru, WANG Haifeng, WU Chuankun
    2020, 40(4):  947-953.  DOI: 10.11772/j.issn.1001-9081.2019111919
    Asbtract ( )   PDF (775KB) ( )  
    References | Related Articles | Metrics
    Smart contract is one of the core technologies of blockchain,and its security and reliability are very important. With the popularization of blockchain application,the number of smart contracts has increased explosively. And the vulnerabilities of smart contracts will bring huge losses to users. However,the current research focuses on the semantic analysis of Ethereum smart contracts,the modeling and optimization of symbolic execution,and does not specifically describe the process of detecting smart contract vulnerabilities using symbolic execution technology,and how to detect common vulnerabilities in smart contracts. Based on the analysis of the operation mechanism and common vulnerabilities of Ethereum smart contract,the symbol execution technology was used to detect vulnerabilities in smart contracts. Firstly,the smart contract control flow graph was constructed based on Ethereum bytecode,then the corresponding constraint conditions were designed according to the characteristics of smart contract vulnerabilities,and the constraint solver was used to generate software test cases to detect the common vulnerabilities of smart contracts such as integer overflow,access control,call injection and reentry attack. The experimental results show that the proposed detection scheme has good detection effect, and has the accuracy of smart contract vulnerability detection up to 85% on 70 smart contracts with vulnerabilities in Awesome-Buggy-ERC20-Tokens.
    Blockchain electronic counting scheme based on practical Byzantine fault tolerance algorithm
    LI Jing, JING Xu, YANG Huijun
    2020, 40(4):  954-960.  DOI: 10.11772/j.issn.1001-9081.2019091559
    Asbtract ( )   PDF (743KB) ( )  
    References | Related Articles | Metrics
    For the problems that third party counting institution does not meet the decentralization and de-trusting characteristics of blockchain and is lack of credibility,a blockchain electronic counting scheme based on the Practical Byzantine Fault Tolerance (PBFT) algorithm was proposed. Firstly,the centerless counting model was built in the distributed environment,and the counting node was determined by the credibility level of the node. Secondly,the consensus of pending ballots was formed based on PBFT. Thirdly,the minimum number of honest nodes in PBFT was set as the threshold for threshold signature,and the threshold signature was only formed by results satisfying the threshold. Finally, the results satisfying the trusted state were recorded in the blockchain account book. Test and analysis results show that only when the honest nodes exceed two-thirds,the PBFT is satisfied,and the obtained counting result is credible.
    Secure storage model of electronic health records based on blockchain
    BAI Yameng, MAN Junfeng, ZHANG Hong
    2020, 40(4):  961-965.  DOI: 10.11772/j.issn.1001-9081.2019081417
    Asbtract ( )   PDF (664KB) ( )  
    References | Related Articles | Metrics
    In order to solve the security risk of electronic health record storage in the context of centralized trust,a secure storage model of electronic health records based on blockchain was proposed. Firstly,a new blockchain storage structure for electronic health records was designed,which solved the problem that the existing storage model was not practical. Then,the workflow of block addition in blockchain was optimized,and the information transmission amount in the Peer to Peer(P2P)network was reduced. Finally,a consensus mechanism based on random number election was proposed to improve the time efficiency of block addition in blockchain. Simulation experiments in Hyperledger were carried out to test the proposed model,and the performance comparison between the proposed consensus mechanism and the existing consensus mechanisms was performed. Experimental results show that the proposed storage model has high security and running time efficiency. This model is suitable for the applications of blockchain in healthcare field and provides a support for the safe sharing of electronic health records.
    Artificial intelligence
    Patent quality evaluation using deep learning with similar papers as augmented dataset
    WEI Wei, LI Xiaojuan
    2020, 40(4):  966-971.  DOI: 10.11772/j.issn.1001-9081.2019091590
    Asbtract ( )   PDF (1017KB) ( )  
    References | Related Articles | Metrics
    In practical application,the patent quality evaluation is usually adopted by experts scoring or the quality evaluation index designed by the experts,so that the evaluation results are subjective and cannot be agreed by the both sides of the evaluation at the same time. In order to solve these problems,a deep learning patent quality evaluation method based on paper similarity calculation was proposed. Firstly,the papers were selected as the objective evaluation data,and the papers were used to calculate the similarity with the patent for augmented data. Then,a deep neural network was introduced to train the quality evaluation model,which was able to realize the map between the similarity of the paper and the quality of the patent to be evaluated. Finally,the quality evaluation model was used to access the patent quality. With perfect score of 100,the simulation results show that in different fields,compared to the corresponding expert evaluation result,the deviation of patent quality evaluation scores obtained by the proposed method is lower than 4,indicating that the proposed method has an effective patent quality evaluation ability.
    Improving machine simultaneous interpretation by punctuation recovery
    CHEN Yuna, SHI Xiaodong
    2020, 40(4):  972-977.  DOI: 10.11772/j.issn.1001-9081.2019101711
    Asbtract ( )   PDF (1373KB) ( )  
    References | Related Articles | Metrics
    In the Machine Simultaneous Interpretation(MSI)pipeline system,semantic incompleteness occurs when the Automatic Speech Recognition(ASR)outputs are directly input into Neural Machine Translation(NMT). To address this problem,a model based on Bidirectional Encoder Representation from Transformers (BERT) and Focal Loss was proposed. Firstly,several segments generated by the ASR system were cached and formed into a string. Then a BERT-based sequence labeling model was used to recover the punctuations of the string,and Focal Loss was used as the loss function in the process of model training to alleviate the class imbalance problem of more unpunctuated samples than punctuated samples. Finally,the punctuation-restored string was input into NMT. Experimental results on English-German and Chinese-English translation show that in term of translation quality,the MSI using the proposed punctuation recovery model has the improvement of 8. 19 BLEU and 4. 24 BLEU respectively compared with the MSI with ASR outputs directly inputting into NMT,and has the improvement of 2. 28 BLEU and 3. 66 BLEU respectively compared with the MSI using punctuation recovery model based on bi-directional recurrent neural network with attention mechanism. Therefore,the proposed model can be effectively applied to MSI.
    Automatic smart contract classification model based on hierarchical attention mechanism and bidirectional long short-term memory neural network
    WU Yuxin, CAI Ting, ZHANG Dabin
    2020, 40(4):  978-984.  DOI: 10.11772/j.issn.1001-9081.2019081327
    Asbtract ( )   PDF (1079KB) ( )  
    References | Related Articles | Metrics
    For that the variety of smart contract applications on the blockchain platform exists more widely and manual filtering the suitable smart contract application services is more difficult, a hierarchical attention mechanism and Bidirectional Long Short-Term Memory(Bi-LSTM)neural network based model was proposed for automatic smart contract classification,namely HANN-SCA(Hierarchical Attention Neural Network with Source Code and Account). Firstly,the Bi-LSTM network was used to simultaneously model the smart contract source code and account information to extract the feature information of smart contract to the greatest extent. The source code perspective focused on the semantic features of code, and the account information perspective focused on the features of the account. Then,in the process of feature learning,the attention mechanism was introduced into the word level and the sentence level respectively to focus on the words and sentences that were important to the classification of smart contract. Finally,the code features and the account features were spliced to generate the document-level feature representation of the smart contract,and the classification task was completed through the Softmax layer. Experimental results on datasets of Dataset-E,Dataset-N and Dataset-EO show that the classification precisions of HANN-SCA model reach 93. 1%,91. 7% and 92. 1% respectively,which are better than those of the traditional Support Vector Machine(SVM)model and other neural network benchmark models,and the proposed model also has better stability and higher convergence speed.
    Image description generation method based on multi-spatial mixed attention
    LIN Xianzao, LIU Jun, TIAN Sheng, XU Xiaokang, JIANG Tao
    2020, 40(4):  985-989.  DOI: 10.11772/j.issn.1001-9081.2019091569
    Asbtract ( )   PDF (800KB) ( )  
    References | Related Articles | Metrics
    Concerning the vacancy of automatic information generation in offshore ship monitoring system,and aiming to build an intelligent ship monitoring system,an image description generation method based on multi-spatial mixed attention was proposed to describe the offshore ship images. The image description generation task is designed to let the computer describe the content of the image with words satisfying linguistics. Firstly,the multi-spatial mixed attention model was trained by the encoding features of the region of interest on the image,then the pretrained decoding model was fine-tuned by reconstructing the loss function with gradient policy,and the final model was obtained. Experimental results on MSCOCO (MicroSoft Common Objects in COntext)image description dataset show that the proposed model is better than the previous attention model on the evaluation index of image description generation,such as CIDEr score. The main content of ship image can be automatically described by the model on the self-constructed ship description dataset,demonstrating that the method can provide the data support for automatic information generation.
    Real-time facial expression and gender recognition based on depthwise separable convolutional neural network
    LIU Shangwang, LIU Chengwei, ZHANG Aili
    2020, 40(4):  990-995.  DOI: 10.11772/j.issn.1001-9081.2019081438
    Asbtract ( )   PDF (1052KB) ( )  
    References | Related Articles | Metrics
    Aiming at the problem of the current common Convolutional Neural Network(CNN)in the expression and gender recognition tasks,that is training process is complicated,time-consuming,and poor in real-time performance,a realtime facial expression and gender recognition model based on depthwise separable convolutional neural network was proposed. Firstly,the Multi-Task Convolutional Neural Network(MTCNN)was used to detect faces in different scale input images,and the detected face positions were tracked by Kernelized Correlation Filter(KCF)to increase the detection speed. Then,the bottleneck layers of convolution kernels of different scales were set,the kernel convolution units were formed by the feature fusion method of channel combination,the diversified features were extracted by the depthwise separable convolutional neural network with residual blocks and separable convolution units,and the number of parameters was reduced to lightweight the model structure. Besides,real-time enabled backpropagation visualization was used to reveal the dynamic changes of the weights and characteristics of learning. Finally,the two networks of expression recognition and gender recognition were combined in parallel to realize real-time recognition of expression and gender. Experimental results show that the proposed network model has a recognition rate of 73. 8% on the FER-2013 dataset,a recognition rate of 96% on the CK+ dataset,the accuracy of gender classification on the IMDB dataset reaches 96%;and this model has the overall processing speed reached 70 frames per second,which is improved by 1. 5 times compared with the method of common convolutional neural network combined with support vector machine. Therefore,for datasets with large differences in quantity,resolution and size,the proposed network model has fast detection,short training time,simple feature extraction, and high recognition rate and real-time performance.
    3D point cloud head pose estimation based on deep learning
    XIAO Shihua, SANG Nan, WANG Xupeng
    2020, 40(4):  996-1001.  DOI: 10.11772/j.issn.1001-9081.2019081479
    Asbtract ( )   PDF (794KB) ( )  
    References | Related Articles | Metrics
    Fast and reliable head pose estimation algorithm is the basis of many high-level face analysis tasks. In order to solve the problem of existing algorithms such as illumination changes,occlusions and large pose variations,a new deep learning framework named HPENet was proposed. Firstly,with the point cloud data used as input,the feature points were extracted from the point cloud structure by using the farthest point sampling algorithm. With feature points as centers,points within spheres with several radiuses were grouped for the further feature description. Then,the multi-layer perceptron and the maximum pooling layer were used to implement the feature extraction of the point cloud,and the predicted head pose was output by the extracted features through the fully connected layer. To verify the effectiveness of HPENet,experiments were carried out on the Biwi Kinect Head Pose dataset. Experimental results show that the errors on angles of pitch,roll and yaw produced by HPENet are 2. 3,1. 5 and 2. 4 degree respectively,and the average time cost of HPENet is 8 ms per frame. Compared with other excellent algorithms,the proposed method has a better performance in terms of both accuracy and computational complexity.
    FaceYoLo algorithm for face detection on mobile platform
    REN Haipei, LI Teng
    2020, 40(4):  1002-1008.  DOI: 10.11772/j.issn.1001-9081.2019091535
    Asbtract ( )   PDF (1482KB) ( )  
    References | Related Articles | Metrics
    Concerning the problem of low real-time performance of face detection on mobile platform,a FaceYoLo real-time face detection algorithm based on deep learning was proposed. Firstly,based on the YoLov3 detection algorithm,the Rapidly Digested Convolutional Layers (RDCL) were added to reduce the input space size, then Multiple Scale Convolutional Layers(MSCL)were added to enrich the receptive fields of different detection scales,and finally the central loss and densification strategy were added to strengthen the generalization ability and robustness of the model. The experimental results show that,when tested on the GPU,the proposed algorithm improves the speed by nearly 8 times compared with the YoLov3 algorithm,has the processing time of each image reached 0. 002 8 s,and increases the accuracy by 2. 1 percentage points;when tested on the Android platform,the proposed algorithm has the detection rate increased from 5 frame/s to 10 frame/s compared with the best MobileNet model,demonstrating that the algorithm can effectively improve the real-time performance of face detection on mobile platform.
    Face liveness detection algorithm based on deep learning and feature fusion
    DENG Xiong, WANG Hongchun
    2020, 40(4):  1009-1015.  DOI: 10.11772/j.issn.1001-9081.2019091595
    Asbtract ( )   PDF (938KB) ( )  
    References | Related Articles | Metrics
    Aiming at the problem that the existing liveness detection algorithms based on deep learning are mostly based on large convolutional neural network,a liveness detection algorithm based on lightweight network MobileNetV2 and feature fusion was proposed. Firstly,the improved MobileNetV2 was used as the basic network to extract features from RGB,HSV and LBP images respectively. Then,the obtained feature maps were stacked together to perform the feature layer fusion. Finally,the features were extracted from the merged feature maps,and the Softmax layer was used to make the judgment whether the face was real or fake. Simulation results show that the Equal Error Rate(EER)of the proposed algorithm on NUAA dataset was 0. 02%,the Average Classification Error Rate(ACER)on Siw dataset was 0. 75%,and the time to test single image costed 6 ms. Experimental results verify that:the fusion of different information can obtain a lower error rate, and the improved lightweight network guarantees the efficiency of the algorithm and meets the real-time requirement.
    Zero-shot image classification based on visual error and semantic attributes
    XU Ge, XIAO Yongqiang, WANG Tao, CHEN Kaizhi, LIAO Xiangwen, WU Yunbing
    2020, 40(4):  1016-1022.  DOI: 10.11772/j.issn.1001-9081.2019081475
    Asbtract ( )   PDF (905KB) ( )  
    References | Related Articles | Metrics
    In the practical applications of image classification,some categories may have no labeled training data at all. The purpose of Zero-Shot Learning(ZSL)is to transfer knowledge such as image features of labeled categories to unlabeled categories and to correctly classify the unlabeled categories. However,the existing state-of-the-art methods cannot explicitly distinguish the input image belonging to the known categories or unknown categories,which leads to a large performance gap for unlabeled categories between the traditional ZSL prediction and the Generalized ZSL(GZSL)prediction. Therefore,a method of fusing of visual error and semantic attributes was proposed to alleviate the prediction bias problem in zero-shot image classification. Firstly,a semi-supervised learning based generative adversarial network framework was designed to obtain visual error information,so as to predict whether the image belongs to the known categories. Then,a zero-shot image classification network combining semantic attributes was proposed to achieve zero-shot image classification. Finally,the performance of zero-shot image classification algorithm combining visual error and semantic attributes was tested on AwA2 (Animal with Attributes) and CUB (Caltech-UCSD-Birds-200-2011) datasets. The experimental results show that, compared to the baseline models,the proposed method can effectively alleviate the prediction bias problem,and has the harmonic index H increased by 31. 7 percentage points on AwA2 dataset and 8. 7 percentage points on CUB dataset.
    Outdoor weather image classification based on feature fusion
    GUO Zhiqing, HU Yongwu, LIU Peng, YANG Jie
    2020, 40(4):  1023-1029.  DOI: 10.11772/j.issn.1001-9081.2019081449
    Asbtract ( )   PDF (1685KB) ( )  
    References | Related Articles | Metrics
    Weather conditions have great influence on the imaging performance of outdoor video equipment. In order to achieve the adaptive adjustment of imaging equipment in inclement weather,so as to improve the effect of intelligent monitoring system,by considering the characteristics that the traditional weather image classification methods have bad classification effect and cannot classify similar weather phenomena,and aiming at the low accuracy of deep learning methods on the weather recognition,a feature fusion model combining traditional methods with deep learning methods was proposed. In the fusion model,four artificially designed algorithms were used to extract traditional features,and AlexNet was used to extract deep features. The eigenvectors after fusion were used to discriminate the image weather conditions. The accuracy of the fusion model on a multi-background dataset reaches 93. 90%,which is better than those of three common methods for comparison,and also performs well on the Average Precision(AP)and Average Recall(AR)indicators;the model has the accuracy on a single background dataset reached 96. 97%,has the AP and AR better than those of other models,and can well recognize weather images with similar features. The experimental results show that the proposed feature fusion model can combine the advantages of traditional methods and deep learning methods to improve the accuracy of existing weather image classification methods,as well as improve the recognition rate under weather phenomena with similar features.
    Semi-supervised hyperspectral image classification based on focal loss
    ZHANG Kailin, YAN Qing, XIA Yi, ZHANG Jun, DING Yun
    2020, 40(4):  1030-1037.  DOI: 10.11772/j.issn.1001-9081.2019081390
    Asbtract ( )   PDF (3567KB) ( )  
    References | Related Articles | Metrics
    Concerning the difficult acquisition of training data in HyperSpectral Image(HSI),a new semi-supervised classification framework for HSI was adopted,in which both limited labeled data and abundant unlabeled data were used to train deep neural networks. At the same time,the unbalanced distribution of hyperspectral samples leads to huge differences in the classification difficulty of different samples,and the original cross-entropy loss function is unable to describe this distribution feature,so the classification effect is not ideal. To address this problem,a multi-classification objective function based on focal loss was proposed in the semi-supervised classification framework. Finally,considering the influence of spatial information of HSI on classification,combined with Markov Random Field(MRF),the sample space features were used to further improve the classification effect. The proposed method was compared with various classical methods on two commonly used HSI datasets. Experimental results show that the proposed method can obtain classification results superior to other comparison methods.
    Intelligent extraction of remote sensing information on large-scale water based on visual attention mechanism
    WANG Quanfang, ZHANG Mengru, ZHANG Yu, WANG Qianqian, CHEN Longyue, YANG Yuqi
    2020, 40(4):  1038-1044.  DOI: 10.11772/j.issn.1001-9081.2019081492
    Asbtract ( )   PDF (2555KB) ( )  
    References | Related Articles | Metrics
    In order to solve the intelligence extraction of information in the era of remote sensing big data,it is important to build the model and method of intelligent information analysis fitting the intrinsic characteristics of remote sensing data. To meet the demand of universal remote sensing intelligent acquisition of large-scale water information,an intelligent extraction method of remote sensing water information based on visual selective attention mechanism and AdaBoost algorithm was proposed. Firstly,by the optimization design of RGB color scheme of remote sensing multi-feature index,the enhancement and visual representation of the water information image features were realized. Then,in HSV color space,the key node information of the chromatic aberration distance image was used to construct the classification feature set,and AdaBoost algorithm was used to construct the water recognition classifier. On this basis,the category that the water belongs to was automatically recognized from the image color clustering result,so as to realize the intelligent extraction of water information. Experimental results show that the proposed method has the water information extraction results improved on Leak Rate(LR)and Composite Classification Accuracy(CCA). At the same time,the proposed method not only effectively reduces the dependence on high quality training samples,but also has good performance on the recognition of temporary water areas such as water with high sediment concentration at wet season and submerged area caused by flooding.
    Intelligent traffic sign recognition method based on capsule network
    CHEN Lichao, ZHENG Jiamin, CAO Jianfang, PAN Lihu, ZHANG Rui
    2020, 40(4):  1045-1049.  DOI: 10.11772/j.issn.1001-9081.2019091610
    Asbtract ( )   PDF (864KB) ( )  
    References | Related Articles | Metrics
    The scalar neurons of convolutional neural networks cannot express the feature location information,and have poor adaptability to the complex vehicle driving environment,resulting in low traffic sign recognition rate. Therefore,an intelligent traffic sign recognition method based on capsule network was proposed. Firstly,the very deep convolutional neural network was used to improve the feature extraction part. Then,a pooling layer was introduced in the main capsule layer. Finally,the movement index average method was used for improving the dynamic routing algorithm. The test results on the GTSRB dataset show that the improved capsule network method improves the recognition accuracy in special scenes by 10. 02 percentage points. Compared with the traditional convolutional neural network,the proposed method has the recognition time for single image decreased by 2. 09 ms. Experimental results show that the improved capsule network method can meet the requirement of accurate and real-time traffic sign recognition.
    Vehicle face recognition algorithm based on NMF with weighted and orthogonal constraints
    WANG Jinkai, JIA Xu
    2020, 40(4):  1050-1055.  DOI: 10.11772/j.issn.1001-9081.2019081338
    Asbtract ( )   PDF (968KB) ( )  
    References | Related Articles | Metrics
    Facing with multi-category samples with limited number of annotations,in order to improve vehicle face recognition accuracy,a vehicle face recognition algorithm based on improved Nonnegative Matrix Factorization(NMF)was proposed. Firstly,the shape feature of local region of vehicle face image was extracted by Histogram of Oriented Gradients (HOG)operator,which was used as the original feature of vehicle face image. Then,the NMF model with multiple weights, orthogonality and sparse constraints was proposed,based on which,the feature bases describing the vehicle face image key regions were acquired,and the feature dimension reduction was achieved. Finally,the discrete cosine distance was used to calculate the similarity between features,and it was able to be concluded that whether the vehicle face images were matched or not. Experimental results show that the proposed recognition algorithm can obtain good recognition effect with accuracy of 97. 68% on the established vehicle face image dataset,at the same time,the proposed algorithm can meet the real-time requirement.
    Bridge crack classification and measurement method based on deep convolutional neural network
    LIANG Xuehui, CHENG Yunze, ZHANG Ruijie, ZHAO Fei
    2020, 40(4):  1056-1061.  DOI: 10.11772/j.issn.1001-9081.2019091546
    Asbtract ( )   PDF (1043KB) ( )  
    References | Related Articles | Metrics
    In order to improve the detection level of bridge cracks,and solve the time-consuming and laborious problem in manual detection and the parameters to be set manually in traditional image processing methods,an improved bridge crack detection algorithm was proposed based on GoogLeNet. Firstly,a large-scale bridge crack Retinex-Laplace-Histogram equalization(RLH)dataset was constructed for model training and testing. Secondly,based on the original GoogLeNet model,the inception module was improved by using the normalized convolution kernel,three improved schemes were used to modify the beginning of the network,the seventh and later inception layers were removed,and a bridge crack feature image classification system was established. Finally,the sliding window was used to accurately locate the cracks and the lengths and widths of the cracks were calculated by the skeleton extraction algorithm. The experimental results show that compared with the original GoogLeNet network,the improve-GoogLeNet network increased the recognition accuracy by 3. 13%, and decreased the training time to the 64. 6% of the original one. In addition,the skeleton extraction algorithm can consider the trend of the crack,calculate the width more accurately,and the maximum width and the average width can be calculated. In summary,the classification and measurement method proposed in this paper have the characteristics of high accuracy,fast speed,accurate positioning and accurate measurement.
    Third-party construction target detection in aerial images of pipeline inspection based on improved YOLOv2 and transfer learning
    CHEN Guihui, YI Xin, LI Zhongbing, QIAN Jiren, CHEN Wu
    2020, 40(4):  1062-1068.  DOI: 10.11772/j.issn.1001-9081.2019081510
    Asbtract ( )   PDF (1309KB) ( )  
    References | Related Articles | Metrics
    Aiming at the few datasets and low detection rate when the traditional target detection algorithm applying to third-party construction target detection and illegally occupied building detection in the aerial images of drone,an aerial image target detection algorithm based on Aerial-YOLOv2 and transfer learning was proposed. Firstly,the trained network combining with data enhancement and transfer learning strategy was used to expand the dataset size,and K-means clustering analysis was used to obtain the number and size of anchor blocks that meet the characteristics of the proposed dataset. Secondly,the adaptive contrast enhancement was used to pre-process the image. Finally,the improved convolution module was proposed to replace the convolution block in YOLOv2,and the feature fusion multi-scale prediction method was combined for target detection. The comparison experiments of different algorithms and training strategies on the aerial images of drone were carried out. Results show that the accuracy and recall rate of the Aerial-YOLOv2 algorithm combined with various training strategies can respectively reach 95% and 91%,and the detection time per image is 14 ms. It can be seen that the algorithm is suitable for the intelligent detection of third-party construction targets and illegally occupied buildings in the aerial images of drone.
    Data science and technology
    Discrimination information method based on consensus and classification for improving document clustering
    WANG Liuyang, YU Yangxin, CHEN Bolun, ZHANG Hui
    2020, 40(4):  1069-1073.  DOI: 10.11772/j.issn.1001-9081.2019091540
    Asbtract ( )   PDF (886KB) ( )  
    References | Related Articles | Metrics
    Different clustering algorithms are used to design their own strategies. However,each technology has certain limitations when it executes a particular dataset. An adequate choice of Discrimination Information Method(DIM)can ensure the document clustering. To solve these problems,a DIM of Document Clustering based on Consensus and Classification (DCCC) was proposed. Firstly,Clustering by DIM (CDIM) was used to solve the generation of initial clustering for dataset,and two initial cluster sets were generated by two different CDIMs. Then,two initial cluster sets were initialized again by different parameter methods,and a consensus was established by using the relationship between the cluster label information,so as to maximize the sum of documents' discrimination number. Finally,Discrimination Text Weight Classification(DTWC)was chosen as text classifier to assign new cluster label to the consensus,the base partitions were altered by training the text classifier,and the final partition was obtained based on the predicted label information. Experiments on 8 network datasets for clustering verification by BCubed's precision and recall index were carried out. Experimental results show that the clustering results of the proposed consensus and classification method are superior to those of comparison methods.
    Fine-grained sentiment classification of film reviews based on ontological features
    HOU Yanhui, DONG Huifang, HAO Min, CUI Xuelian
    2020, 40(4):  1074-1078.  DOI: 10.11772/j.issn.1001-9081.2019081426
    Asbtract ( )   PDF (588KB) ( )  
    References | Related Articles | Metrics
    In view of the lack of feature attributes and the granularity division on emotion intensity level in Chinese film reviews,a fine-grained sentiment classification model based on ontological features was proposed. Firstly,Term Frequency-Inverse Document Frequency(TF-IDF)and TextRank algorithm were used to extract movie features and construct ontology conceptual model. Secondly,the film attributes and Plutchik's Wheel of Emotion were combined with Bidirectional Long Short-Term Memory (Bi-LSTM) neural network to build a fine-grained emotion classification model based on feature granularity level and eight-category emotion intensity. In the experiments,the analysis of ontological features shows that the movie viewers pay the most attention to the attributes of the story,followed by the features of theme,character,scene and director;Model performance analysis shows that,based on feature granularity and eight-category emotion intensity, compared with other five classification models using emotion dictionary,machine learning and Bi-LSTM network algorithm at the level of overall granularity and three-category emotion intensity,the proposed model not only has a higher F1 value (0. 93),but also can provide viewers with a reference to emotional preferences and emotional intensities of film attributes, and achieves a more fine-grained emotional classification of Chinese film reviews.
    Self-driving tour route mining based on sparse trajectory clustering
    YANG Fengyi, MA Yupeng, BAO Hengbin, HAN Yunfei, MA Bo
    2020, 40(4):  1079-1084.  DOI: 10.11772/j.issn.1001-9081.2019081467
    Asbtract ( )   PDF (1419KB) ( )  
    References | Related Articles | Metrics
    Aiming at the difficulty of constructing real tour routes from sparse refueling trajectories of self-driving tourists,a sparse trajectory clustering algorithm based on semantic representation was proposed to mine popular self-driving tour routes. Different from traditional trajectory clustering algorithms based on trajectory point matching,in this algorithm, the semantic relationships between different trajectory points were considered and the low-dimensional vector representation of the trajectory was learned. Firstly,the neural network language model was used to learn the distributed vector representation of the gas stations. Then,the average value of all the station vectors in each trajectory was taken as the vector representation of this trajectory. Finally,the classical k-means algorithm was used to cluster the trajectory vectors. The final visualization results show that the proposed algorithm mines two popular self-driving tour routes effectively.
    Cyber security
    Response obfuscation based secure deduplication method for cloud data with resistance against appending chunk attack
    TANG Xin, ZHOU Linna
    2020, 40(4):  1085-1090.  DOI: 10.11772/j.issn.1001-9081.2019081468
    Asbtract ( )   PDF (901KB) ( )  
    References | Related Articles | Metrics
    Appending chunk attack is an important attack to threaten the security of cross-user deduplication for cloud data,which works by appending a random number of non-duplicate chunks to the file to be detected,making it impossible for cloud service providers to determine the true existence of the file. Therefore,the existence privacy of cloud data cannot be protected by general ways of response obfuscation methods. To deal with this problem,a new response obfuscation based secure deduplication method with resistance against appending chunk attack was proposed. By calculating the number of appending chunks,counting the number of non-duplicate chunks and comparing these two to determine the minimum number of redundant chunks involved in the response,so as to achieve the obfuscation. As a result,the existence of the checking file was not able to be judged by the attacker according to the response with little extra communication overhead. Security analysis and experimental results show that,compared with the state-of-the-art in this field,the proposed method achieves higher level of security with smaller amount of overhead required,or improves security significantly with comparable or slightly increased overhead.
    Advanced computing
    Surrogate-based differential evolution constrained optimization
    XUE Feng, SHI Xuhua, SHI Feifan
    2020, 40(4):  1091-1096.  DOI: 10.11772/j.issn.1001-9081.2019091587
    Asbtract ( )   PDF (641KB) ( )  
    References | Related Articles | Metrics
    In order to solve the constrained optimization problem of objective function with time-consuming computation,a surrogate model was proposed to replace the objective function with time-consuming computation,the constraint individuals were selected based on the information of the objective function,and a surrogate-based differential evolution constrained optimization algorithm was proposed. Firstly,the Latin hypercube sampling method was used to establish the initial population,which was evaluated by the objective function with time-consuming calculation,and the neural network surrogate model of the objective function was established based on these sample data. Then,the differential evolution method was used to generate offsprings for each parent in the population,and the offspring individuals were evaluated by using the surrogate model. The feasibility rule was used to compare the offsprings with their parents and update the population,and the inferior individuals in the population were replaced with the better individuals in the alternate archive according to the replacement mechanism. Finally,the algorithm stopped when the maximum fitness evaluation number was reached,and the optimal solution was obtained. The results of this algorithm and comparison algorithms running on 10 test functions show that the results obtained by this algorithm are more accurate. The results of applying this algorithm to the I-beam optimization problem show that the number of fitness evaluations of this algorithm is reduced by 80% compared with that of the algorithm before optimization,and the number of fitness evaluations of this algorithm is reduced by 36% compared with that of FROFI (Feasibility Rule with the incorporation of Objective Function Information) algorithm. Using the proposed algorithm to realize the optimization can effectively reduce the number of calls for the objective function with time-consuming computation,improve optimization efficiency,and save computational cost.
    Modeling and memetic algorithm for vehicle routing problem with simultaneous pickup-delivery and time windows
    ZHANG Qinghua, WU Guangpu
    2020, 40(4):  1097-1103.  DOI: 10.11772/j.issn.1001-9081.2019081355
    Asbtract ( )   PDF (1187KB) ( )  
    References | Related Articles | Metrics
    In order to solve the Vehicle Routing Problem with Simultaneous Pickup-Delivery and Time Windows (VRPSPDTW)in the context of reverse logistics,the corresponding vehicle routing problem model was established according to the actual situation and solved by memetic algorithm. In the process of solving the model,the Guided Ejection Search (GES)was used to generate the initial population. In the process of population evolution,the Edge Assembly Crossover (EAX)method was used to generate the offspring,and in order to improve the quality of solutions and the search efficiency of algorithms,multiple neighborhood structures were used to repair and educate the offspring. The performance of memetic algorithm was tested and compared with Genetic Algorithm (GA),parallel-Simulate Annealing algorithm (p-SA) and Discrete Cuckoo Search(DCS)algorithm on Wang and Chen test dataset. Experimental results show that the proposed algorithm obtains the current optimal solutions when solving all small-scale examples;the algorithm updates or achieves current optimal solutions on 70% examples when solving the standard-scale examples,and the obtained optimal solution has more than 5% improvement compared with the current optimal solution,fully verifying the good performance of the algorithm for solving VRPSPDTW.
    Stochastic support selection based generalized orthogonal matching pursuit algorithm
    XU Zhiqiang, JIANG Tiegang, YANG Libo
    2020, 40(4):  1104-1108.  DOI: 10.11772/j.issn.1001-9081.2019091576
    Asbtract ( )   PDF (797KB) ( )  
    References | Related Articles | Metrics
    Aiming at the problems of high complexity and long reconstruction time of Generalized Orthogonal Matching Pursuit(GOMP)algorithm,a Stochastic support selection based GOMP(StoGOMP)algorithm was proposed. Firstly,the strategy of stochastic support selection was introduced,and a probability value was randomly generated in each iteration. Then the generated probability value was compared to the preset probability value to determine the selection method of candidate support set. If this probability value was less than the preset probability value,the matching calculation method was adopted,otherwise,the random selection method was adopted. Finally,the residual was updated according to the obtained candidate supports. In this way,the balance between the complexity of the single iteration and the number of iterations of the algorithm was fully considered,and the computational cost of the algorithm was reduced. The experiment of one-dimensional random signal reconstruction shows that the number of samples required for StoGOMP algorithm to achieve 100% reconstruction success rate is reduced by 9. 5% compared with that for GOMP algorithm when the preset probability is 0. 5 and the sparsity is 20. The actual image reconstruction experiment shows that the proposed algorithm has the same reconstruction accuracy as GOMP algorithm,and the reconstruction time of the proposed algorithm is reduced by more than 27% compared to that of the original algorithm when the sampling rate is 0. 5,which indicates that StoGOMP algorithm can effectively reduce the signal reconstruction time.
    Network and communications
    Structural signature extraction method for mobile application recognition
    SHEN Liang, WANG Xin, CHEN Shuhui
    2020, 40(4):  1109-1114.  DOI: 10.11772/j.issn.1001-9081.2019081380
    Asbtract ( )   PDF (1236KB) ( )  
    References | Related Articles | Metrics
    Focusing on the needs of mobile application traffic monitoring and behavior analysis,a Hyper Text Transfer Protocol(HTTP)traffic structural signature extraction method was proposed to effectively identify the application to which mobile network traffic belongs. A self-developed Virtual Private Network(VPN)-based traffic collection tool was used to obtain the research data,which was able to accurately identify the application that each data stream belongs to. In the signature extraction stage,the signature composition was not pre-designed,and the structural signatures of the HTTP traffic were obtained through three steps of flow clustering,obtaining the longest common subsequence and character substitution. The signatures of 42 applications were extracted from 117 772 HTTP traffic to identify 50 387 HTTP traffic in test set. The proposed algorithm has the average accuracy of 99%,the average recall of 90. 63%,and the maximum false positive rate of single application of 0. 52%. The experimental results show that the proposed structural signature extraction method can effectively identify the traffic of mobile applications.
    Energy efficiency optimization of heterogeneous cellular networks based on transmitting power of pico base station
    CHEN Yonghong, GUO Lili, ZHANG Shibing, YANG Jie
    2020, 40(4):  1115-1118.  DOI: 10.11772/j.issn.1001-9081.2019071236
    Asbtract ( )   PDF (606KB) ( )  
    References | Related Articles | Metrics
    Energy efficiency of Heterogeneous cellular Network(HetNet)has attracted wide attention in recent years. However,researches on the energy efficiency of macro base stations using non-Poisson process are not enough. To solve this problem,the energy efficiency of the two-tier HetNet was investigated,in which the deployment of macro base stations was modeled by β-Ginibre Point Process(β-GPP). Firstly,a simple approximation method was used to analyze the Signal to Interference Ratio(SIR)distribution in two-tier HetNet,then the coverage probability,the average achievable throughput and the energy efficiency of the system were derived. Finally,an energy efficiency optimization algorithm was proposed to find the optimal transmitting power of pico base station,maximizing the energy efficiency. The simulation results show that when β=1,the distribution density of macro base station is 2×10-4 m-2,and the distribution density of pico base station is 2 times of that of macro base station,the proposed energy efficiency optimization scheme can improve the system energy efficiency by about 20%. The experimental results verify the accuracy of the theoretical analysis and the effectiveness of the proposed energy efficiency optimization algorithm.
    Rate adaption algorithm for embedded multi-channel wireless video transmission
    LUO Chiwei, QU Tao, DENG Dexiang
    2020, 40(4):  1119-1126.  DOI: 10.11772/j.issn.1001-9081.2019081503
    Asbtract ( )   PDF (833KB) ( )  
    References | Related Articles | Metrics
    Wireless video transmission and video compression technology are the foundations and cores of many Internet of Things(IoT)applications and embedded systems in these days. However,multi-channel transmission always causes video frame loss and delay jitter because of the continuous change of wireless network state. Although the adaption algorithm can solve the video transmission problem under PC or server platform to a certain extent,the real-time performance and Quality of Service(QoS)requirement cannot be satisfied under the embedded platform and wireless network. Therefore, based on the DM368 chip,a complete platform was designed from video capture,compression,WiFi transmission,control unit reception to host computer display. At the same time,with the full consideration of the characteristics of embedded platform,a rate adaption algorithm that combines signal quality,network bandwidth,buffer status and congestion control was proposed. In this algorithm,the Gaussian function was used to calculate network bandwidth,the segmented inverse proportional function was used to adjust buffer status,the weighted moving method was adopted to smooth rate,and the extreme value suppression method was used for rate balancing. The smooth rate adjustment was realized by this algorithm, and the algorithm was applied to the proposed platform to realize the management of the control unit on multiple WiFi cameras,multi-channel transmission and load balancing. The QoS was used as the evaluation index for experimental verification. The results show that the algorithm performs well on the embedded platform with great improvements of smoothness and buffer stability,and has significantly fairness and bandwidth utilization improvements under multi-channel condition. In a variety of situations,such as single camera signal quality dynamic change or multi-camera bandwidth competition,compared with the McGinely Dynamic Indicator(MDI)algorithm,the proposed algorithm has the smoothness improved by 16% to 59%;compared with the Buffer-Based Algorithm(BBA),the proposed algorithm has the cache jitter reduced by 15% to 72%,and the delay jitter reduced by 12% to 76%.
    Computer software technology
    Domain model of Web-based dynamic geometry software and its applications
    GUAN Hao, QIN Xiaolin, RAO Yongsheng, CAO Sheng
    2020, 40(4):  1127-1132.  DOI: 10.11772/j.issn.1001-9081.2019091672
    Asbtract ( )   PDF (1285KB) ( )  
    References | Related Articles | Metrics
    Dynamic geometry software is widely applied to geometric constraint constructions because it is dynamic and intuitive. Aiming at a problem that the data structures in the field of dynamic geometry lack reusable abstract descriptions,a design method of the dynamic geometric software domain model was proposed. Firstly,the basic context boundaries were identified and outlined by means of domain analysis. Then,a dynamic geometry software core domain model was designed through the domain model. Finally,the dynamic geometry software was decoupled in both vertical and horizontal dimensions during the architecture modeling process. Experimental results show that the dynamic geometry software developed by using the design method of the proposed domain model can correctly deal with the graphic degradation situation at a critical position. The domain knowledge expressed by the model is applicable to 2D and 3D dynamic geometry software at the same time,and can design the layout and interaction for different devices respectively,thus a high-level reuse of the domain knowledge is achieved.
    Virtual reality and multimedia computing
    Local feature point matching algorithm with anti-affine property
    QIU Yunfei, LIU Xing
    2020, 40(4):  1133-1137.  DOI: 10.11772/j.issn.1001-9081.2019091588
    Asbtract ( )   PDF (635KB) ( )  
    References | Related Articles | Metrics
    In order to solve the problems that the existing local feature matching algorithm has poor matching effect and high time cost on affine images,and RANdom SAmple Consensus(RANSAC)algorithm cannot obtain a good parameter model on affine image matching,Affine Accelerated KAZE(A-AKAZE)algorithm with anti-affine property was proposed and the vector field consistency was used to screen interior points. Firstly,the scale space was constructed by using the nonlinear function,then the feature points were detected by Hessian matrix,and the appropriate areas were selected as the feature sampling windows with the feature points as the centers. Secondly,the feature sampling windows were projected on longitude and latitude to simulate the influence of different angles on the image,and then the Affine Modified-Local Difference Binary(A-MLDB)descriptors with anti-affine property were extracted from the projection region. Finally,the interior points were extracted by the vector field consistency algorithm. Experimental results show that the correct matching rate of A-AKAZE algorithm is more than 20% higher than that of AKAZE algorithm,is about 15% higher than that of AKAZE+RANSAC algorithm,is about 10% higher than that of Affine Scale-Invariant Feature Transform(ASIFT)algorithm, and is 5% higher than that of ASIFT+RANSAC algorithm;at the same time,A-AKAZE algorithm has the matching speed much higher than AKAZE+RANSAC,ASIFT and ASIFT+RANSAC algorithms.
    Image registration algorithm combining GMS and VCS+GC-RANSAC
    DING Hui, LI Lihong, YUAN Gang
    2020, 40(4):  1138-1143.  DOI: 10.11772/j.issn.1001-9081.2019081465
    Asbtract ( )   PDF (1935KB) ( )  
    References | Related Articles | Metrics
    Aiming at the problems of long registration time and low registration accuracy of current image registration algorithms,an image registration algorithm based on Grid-based Motion Statistics(GMS),Vector Coefficient Similarity (VCS)and Graph-Cut RANdom SAmple Consensus(GC-RANSAC)was proposed. Firstly,the feature points of the image were extracted through the ORB(Oriented FAST and Rotated BRIEF)algorithm,and Brute-Force matching of the feature points was performed. Then,the coarse matching feature points in the image were meshed by the GMS algorithm,and the coarse matching pairs were filtered based on the principle that high feature support exists in the neighborhood of the correct matching points in the grid. And the part elimination was performed to the matching pairs by introducing the principle that the image matching pair has VCS not exceed a set threshold during vector operation,which is beneficial to the fast convergence of the algorithm in the later stage. Finally,the local optimal model fitting was performed by using the GC-RANSAC algorithm to obtain the fine matching feature point set and achieve image registration and stitching with high precision. Compared with algorithms such as ASIFT+RANSAC,GMS,AKAZE+RANSAC,GMS+GC-RANSAC,the results show that the proposed algorithm improves the average matching accuracy by 30. 34% and reduces the average matching time by 0. 54 s.
    Component contour modification method based on image registration
    WU Menghua, HU Xiaobing, LI Hang, JIANG Daiyu
    2020, 40(4):  1144-1150.  DOI: 10.11772/j.issn.1001-9081.2019081463
    Asbtract ( )   PDF (960KB) ( )  
    References | Related Articles | Metrics
    Focusing on the problem that the component contours taken by smart machine tool visual system always contain abnormal regions caused by the background interference,a component contour modification method based on image registration was proposed. Firstly,the component template feature point set and the matched feature point set were extracted from the component engineering drawing and the real image. Secondly,the parameters in the affine transformation model were decomposed and analyzed,and a criterion function was established based on area characteristics and edge structure characteristics of feature point sets of both images. Thirdly,an improved genetic algorithm was used to search for affine transformation parameters corresponding to the global maximum similarity between two images. After the image registration, the abnormal contour segments were detected and replaced by calculating the optimal migrated piecewise Hausdorff distance between the template contour point set and the matched contour point set. Experimental results show that the proposed method can detect the abnormal contour segments in matching contour point set with high accuracy and stability,its registration accuracy is 50% higher than that of Square Summation Joint Feature(SSJF)method,and the distance where the modified contour intersects is less than 3 pixels.
    Intrinsic shape signature algorithm based on adaptive neighborhood
    SHI Zhiliang, CAI Wangyue, WANG Guoqiang, XIONG Linjie
    2020, 40(4):  1151-1156.  DOI: 10.11772/j.issn.1001-9081.2019091538
    Asbtract ( )   PDF (5956KB) ( )  
    References | Related Articles | Metrics
    Concerning the problems that the determination of fixed scale in 3D point cloud feature point detection algorithm requires empirical knowledge,and the calculation of adaptive scale requires much time cost,an improved algorithm named Adaptive Neighborhood Intrinsic Shape Signature(ANISS)was proposed. Firstly,the local features were used to calculate the adaptive neighborhood k value of each point. Then,the k value was used as the neighborhood size of the ANISS algorithm,and by comparing the ratio of the continuous eigenvalues with the threshold,the approximate feature points were obtained. Finally,the k values of the approximate feature points were used as the neighborhood size of the Non-Maximum Suppression(NMS),and the NMS algorithm was executed to obtain the final feature points. The results of rotational translation invariance experiment and noise sensitivity experiment show that the repeatability of the feature points detected by ANISS algorithm is higher than that of Intrinsic Shape Signature(ISS)algorithm. ANISS algorithm not only reduces the inaccuracy caused by the neighborhood parameter input in ISS algorithm,but also has high computational efficiency.
    Real-time SLAM algorithm with keyframes determined by inertial measurement unit
    WEI Wenle, JIN Guodong, TAN Lining, LU Libin, CHEN Danqi
    2020, 40(4):  1157-1163.  DOI: 10.11772/j.issn.1001-9081.2019081326
    Asbtract ( )   PDF (3649KB) ( )  
    References | Related Articles | Metrics
    Due to the limitation of the computational power of embedded processors,the poor real-time performance has always been an urgent problem to be solved in the practical applications of Visual Inertial Simultaneous Localization And Mapping(VI-SLAM). Therefore,a real-time Simultaneous Localization And Mapping(SLAM)with keyframes determined by Inertial Measurement Unit(IMU)was proposed,which was mainly divided into three threads:tracking,local mapping and loop closing. Firstly,the keyframes were determined adaptively by the tracking thread through the IMU pre-integration, and the adaptive threshold was derived from the result of the visual inertia tight coupling optimization. Then,only the keyframes were tracked,thereby avoiding the feature processing to all frames. Finally,a more accurate Unmanned Aerial Vehicle(UAV)pose was obtained by the local mapping thread through the visual inertial bundle adjustment in the sliding window,and the globally consistent trajectory and map were output by the loop closing thread. Experimental results on the dataset EuRoC show that the algorithm can significantly reduce the tracking thread time consumption without loss of precision and robustness,and reduce the dependence of VI-SLAM on computing resources. In the actual flight test,the true trajectory of the drone with scale information can be estimated accurately by the algorithm in real time.
    Adaptive enhancement algorithm of low illumination image based on maximum difference image decision
    WANG Ruiyao, YUE Xueting, ZHOU Zhiqing, GENG Zexun
    2020, 40(4):  1164-1170.  DOI: 10.11772/j.issn.1001-9081.2019091541
    Asbtract ( )   PDF (1501KB) ( )  
    References | Related Articles | Metrics
    When applying traditional image enhancement algorithm to low illumination images with uneven illumination distribution,it is easy to produce color distortion and over enhancement of bright areas. To resolve theses problems,an adaptive enhancement algorithm of low illumination image based on maximum difference image was proposed. Firstly,the concept of maximum difference image was proposed,and the initial illumination component was roughly estimated by the maximum difference image. Secondly,the method of alternating guided filtering was proposed,which was used to correct the initial illumination component,so as to realize the accurate estimation of illumination component. Finally,the Gamma transform was designed for image brightness adaptivity,which was able to adaptively adjust the Gamma transform parameters according to the acquired illumination components,thus,the influence of uneven illumination was eliminated while enhancing the image. Experimental results show that the enhanced image effectively eliminates the influence of uneven illumination distribution,the brightness,contrast,detail performance and color fidelity of the image are significantly improved,the average gradient increases by more than one time,and the information entropy increases by more than 14%. Because the proposed algorithm estimates the light component accurately,and the adaptive Gamma transform is optimized for low illumination images,so that the proposed algorithm has very effective enhancement effect for color images under weak light conditions like night.
    Mixed non-convex and non-smooth regularization constraint based blind image restoration
    GENG Yuanqian, WU Chuansheng, LIU Wen
    2020, 40(4):  1171-1176.  DOI: 10.11772/j.issn.1001-9081.2019091647
    Asbtract ( )   PDF (2187KB) ( )  
    References | Related Articles | Metrics
    In order to restore high-quality clear images,a regularization constraint based blind image restoration method was proposed. Firstly,in order to improve the accuracy of blur kernel estimation,the regularization term of L0-norm was used to perform sparsity constraint to the blur kernel according to the sparsity of blur kernel. Secondly,in order to retain the edge information of the image,the L0-norm of combining both first and second order image gradients was used to perform regularized constraint to the image gradient according to the sparsity of image gradient. Finally,since the proposed mixed regularization constraint model is essentially a non-convex and non-smooth optimization problem,the model was solved by the alternating direction method of multipliers,and the clear image was restored by using the L1-nom data fitting term and total variation method in the non-blind deconvolution stage. Experimental results show that the proposed method can restore clearer details and edge information,and has higher quality of restoration result.
    Regional-content-aware nuclear norm for low-does CT image denosing
    SONG Yun, ZHANG Yuanke, LU Hongbing, XING Yuxiang, MA Jianhua
    2020, 40(4):  1177-1183.  DOI: 10.11772/j.issn.1001-9081.2019091592
    Asbtract ( )   PDF (5420KB) ( )  
    References | Related Articles | Metrics
    The low-rank constraint model based on traditional Nuclear Norm Minimization(NNM)tends to cause local texture detail loss in the denoising of Low-Dose CT(LDCT)image. To tackle this issue,a regional-content-aware weighted NNM algorithm was proposed for LDCT image denoising. Firstly,a Singular Value Decomposition(SVD)based method was proposed to estimate the local noise intensity in LDCT image. Then,the target image block matching was performed based on the local statistical characteristics. Finally,the weights of the nuclear norms were adaptively set based on both the local noise intensity of the image and the different singular value levels,and the weighted NNM based LDCT image denoising was realized. The simulation results illustrated that the proposed algorithm decreased the Root Mean Square Error(RMSE)index by 30. 11%,14. 38% and 8. 75% respectively compared with the traditional NNM,total variation minimization and transform learning algorithms,and improved the Structural SIMilarity(SSIM)index by 34. 24%,23. 06% and 11. 52% respectively compared with the above three algorithms. The experimental results on real clinical data illustrated that the mean value of the radiologists' scores of the results obtained by the proposed algorithm was 8. 94,which is only 0. 21 lower than that of the corresponding full dose CT images,and was significantly higher than those of the traditional NNM,total variation minimization and transform learning algorithms. The simulation and clinical experimental results indicate that the proposed algorithm can effectively reduce the artifact noise while preserving the texture detail information in LDCT images.
    No-reference image quality assessment method for facial beautification image
    ZHANG Junsheng, XU Jingjing, YU Wei
    2020, 40(4):  1184-1190.  DOI: 10.11772/j.issn.1001-9081.2019091552
    Asbtract ( )   PDF (1179KB) ( )  
    References | Related Articles | Metrics
    In view of the fact that facial beautification has been widely studied,but the lack of effective beautification image quality evaluation methods limits the further development of beautification technology,a no-reference evaluation method for facial beautification image quality was proposed. In this method,the facial cognition and perception were combined with the facial beautification technologies to unearth the quality representation of beautified images. Firstly,a facial beautification image database was constructed,the facial image was decomposed to three areas:skin,eyes and mouth. Then,facial aesthetic features were extracted from five aspects:skin color,smoothness,illumination,grayscale difference and sharpness. Finally,Support Vector Regression(SVR)was used to train the facial beautification quality model and predict the quality of the beautified image. The experimental results show that the proposed method achieves 0. 920 5 and 0. 900 9 respectively in the Pearson linear correlation coefficient and Spearman RankOrder Correlation Coefficient(SROCC) on the proposed database,which are higher than those of image quality evaluation methods BIQI(Blind Image Quality Indices),and NIQE(Natural Image Quality Evaluation).
    Fast echo cancellation algorithm in smart speaker
    ZHANG Wei, WANG Dongxia, YU Ling
    2020, 40(4):  1191-1195.  DOI: 10.11772/j.issn.1001-9081.2019081482
    Asbtract ( )   PDF (2168KB) ( )  
    References | Related Articles | Metrics
    Considering that the microphone array is mostly used as a sound pickup device in the smart speaker,and there are distortion and complexity of the acoustic echo cancellation in the adaptive filtering technology on single channel,a fast echo cancellation algorithm for microphone array was proposed. First of all,the adaptive filtering technology was used to estimate the first channel echo,then estimate the relative echo transfer function between the arrays,and the echoes of other channels were obtained by multiplying the above two. Secondly,the estimated echo and the noise were regarded as the noise reference signals of the Generalized Sidelobe Canceller(GSC)beamforming lower branch,which were removed by the GSC beamforming algorithm. The simulation results show that the proposed algorithm has good echo cancellation and noise suppression performance under moderate reverberation,long distance,low echo to noise ratio and using music as echo environment. And the algorithm not only has small computational complexity,but also makes target speech signals have high signal distortion ratio and intelligibility.
    Improved fuzzy c-means MRI segmentation based on neighborhood information
    WANG Yan, HE Hongke
    2020, 40(4):  1196-1201.  DOI: 10.11772/j.issn.1001-9081.2019091539
    Asbtract ( )   PDF (1675KB) ( )  
    References | Related Articles | Metrics
    In the segmentation of brain image,the image quality is often reduced due to the influence of noise or outliers. And traditional fuzzy clustering has some limitations and is easily affected by the initial value,which brings great trouble for doctors to accurately identify and extract brain tissue. Aiming at these problems,an improved fuzzy clustering image segmentation method based on neighborhoods of image pixels constructed by Markov model was proposed. Firstly,the initial clustering center was determined by Genetic Algorithm(AG). Secondly,the expression of the target function was changed,the calculation method of the membership matrix was changed by adding the correction term in the target function and was adjusted by the constraint coefficient. Finally,the Markov Random Field(WRF)was used to represent the label information of the neighborhood pixels,and the maximized conditional probability of Markov random field was used to represent the neighborhood of the pixel,which improves the noise immunity. Experimental results show that the proposed method has good noise immunity,it can reduce the false segmentation rate and has high segmentation accuracy when used to segment brain images. The average accuracy of the segmented image has Jaccard Similarity(JS)index of 82. 76%,Dice index of 90. 45%,and Sensitivity index of 90. 19%. At the same time,the segmentation of brain image boundaries is clearer and the segmented image is closer to the standard segmentation image.
    Region division method of brain slice image based on deep learning
    WANG Songwei, ZHAO Qiuyang, WANG Yuhang, RAO Xiaoping
    2020, 40(4):  1202-1208.  DOI: 10.11772/j.issn.1001-9081.2019091521
    Asbtract ( )   PDF (3502KB) ( )  
    References | Related Articles | Metrics
    Aiming at the problem of poor accuracy of automatic region division of mouse brain slice image using traditional multimodal registration method,an unsupervised multimodal region division method of brain slice image was proposed. Firstly,based on the mouse brain map,the Atlas brain map and the Average Template brain map in the Allen Reference Atlases (ARA) database corresponding to the brain slice region division were obtained. Then the Average Template brain map and the mouse brain slices were pre-registered and modal transformed by affine transformation preprocessing and Principal Component Analysis Net-based Structural Representation(PCANet-SR)network processing. After that,according to U-net and the spatial transformation network,the unsupervised registration was realized,and the registration deformation relationship was applied to the Atlas brain map. Finally,the edge contour of the Atlas brain map extracted by the registration deformation was merged with the original mouse brain slices in order to realize the region division of the brain slice image. Compared with the existing PCANet-SR+B spline registration method,experimental results show that the Root Mean Square Error(RMSE)of the registration accuracy index of this method reduced by 1. 6%,the Correlation Coefficient(CC)and the Mutual Information(MI)increased by 3. 5% and 0. 78% respectively. The proposed method can quickly realize the unsupervised multimodal registration task of the brain slice image,and make the brain slice regions be divided accurately.
    Frontier & interdisciplinary applications
    Dynamic reinforcement model for driving safety based on cooperative feedback control in Internet of vehicles
    HUANG Chen, CAO Jiannong, WANG Shihui, ZHANG Yan
    2020, 40(4):  1209-1214.  DOI: 10.11772/j.issn.1001-9081.2019101808
    Asbtract ( )   PDF (2663KB) ( )  
    References | Related Articles | Metrics
    In Internet of Vehicles(IoV)environment,a single vehicle cannot meet all the time-sensitive driving safety requirements because of limited capability on information acquiring and processing. Cooperation among vehicles to enhance information sharing and channel access ability is inevitable. In order to solve these problems,a cooperative feedback control algorithm based dynamic reinforcement model for driving safety was proposed. Firstly,a virtual fleet cooperation model was proposed to improve the precision and expand the range of global traffic sensing,and a stable cooperation relationship was constructed among vehicles to form cooperative virtual fleet while avoiding channel congestion. Then,a joint optimization model focusing on message transmission and driving control was implemented,and the deep fusion of heterogeneous traffic data was used to maximize the safety utility of IoV. Finally,an adaptive feedback control model was proposed according to the prediction on spatial-temporal change of traffic flow,and the driving safety strategy was able to be adjusted in real-time. Simulation results demonstrate that the proposed model can obtain good performance indexes under different traffic flow distribution models, can effectively support driving assisted control system, and reduce channel congestion while maintaining driving safety.
    Impact assessment of engineering change propagation for complex products based on multiplex network
    LI Congdong, ZHANG Zhiwei, CAO Cejun, ZHANG Fanshun
    2020, 40(4):  1215-1222.  DOI: 10.11772/j.issn.1001-9081.2019101779
    Asbtract ( )   PDF (3619KB) ( )  
    References | Related Articles | Metrics
    In the process of traditional impact assessment of engineering change propagation for complex products,the joint effect of different association relationships among parts on change propagation is not comprehensively considered. Therefore,the multiplex network theory was applied to the modeling of complex product. Firstly,to explore the impact of different association relationships on engineering change propagation,the parts of a complex product and their functional, behavioral and structural association relationships were abstracted to a multiplex network. Then,the improved all-around distance index was used for evaluating the importance degrees of nodes in the multiplex network. Finally,the Susceptibility-Infection-Susceptibility(SIS)model was used to quantitatively assess the impact of engineering change propagation. To verify the effectiveness of the proposed method,the impact assessment of engineering change propagation of a type of diesel engine was taken as an example. Results demonstrate that:the multiplex network can fully describe complex products with different types of association relationships;the improved all-around distance index increases the ability to distinguish the importance degrees of nodes with the same coreness;the change in high-important parts will cause the avalanche propagation of engineering change throughout the product range.
    Optimization of quay crane assignment based on ship efficiency
    MAO Minli, LIANG Chengji, HU Xiaoyuan
    2020, 40(4):  1223-1230.  DOI: 10.11772/j.issn.1001-9081.2019081528
    Asbtract ( )   PDF (1246KB) ( )  
    References | Related Articles | Metrics
    In the container terminal system,the effective quay crane assignment for vessels is helpful to ease the strain that berths and quay cranes are in short supply in container terminals and improve the operational efficiency of ports. Aiming at the integrated optimization problem of berth allocation and quay crane assignment of dynamic arriving vessels under continuous berth,the quay crane assignment for vessels was dynamically adjusted based on ship efficiency,a model with the purpose of minimizing the total cost containing delayed berthing cost,preference deviated berthing cost,delayed departure cost and quay crane reassignment cost was established,and a heuristic algorithm based on the adjustment rules of quay crane assignment was designed and Genetic Algorithm(GA) was used to solve the model. Finally,the experimental results verified the effectiveness of the proposed model and algorithm in solving the problem of berth allocation and quay crane assignment in actual ports,and by comparing with the results calculated by the traditional GA,the optimization effect of the proposed algorithm was proved.
    Reconstruction of porous media using adaptive deep transfer learning
    CHEN Jie, ZHANG Ting, DU Yi
    2020, 40(4):  1231-1236.  DOI: 10.11772/j.issn.1001-9081.2019091608
    Asbtract ( )   PDF (979KB) ( )  
    References | Related Articles | Metrics
    Aiming at the low efficiency and the complex simulation process of the traditional reconstruction methods for porous media such as Multi-Point Statistics(MPS)which require scanning the training image many times and to obtain simulation results by complex probability calculations,a method to reconstruct porous media using adaptive deep transfer learning was presented. Firstly,deep neural network was used to extract the complex features from the training image of porous media. Secondly,the adaptive layer was added in deep transfer learning to reduce the difference in data distribution between training data and prediction data. Finally,through copying features by transfer learning,the simulation result consistent with the real training data was obtained. The performance of the proposed method was evaluated by comparing with the classical porous media reconstruction method MPS in multiple-point connectivity curve,variogram curve and porosity. The results indicate that the proposed method has high reconstruction quality. Meanwhile,the method has the average running time reduced from 840 s to 166 s,the average CPU usage dropped from 98% to 20%,and the average memory utilization decreased by 69%. The proposed method significantly improves the efficiency of porous media reconstruction under the premise of ensuring better quality of reconstruction results.
    Design and implementation of SIFT algorithm for UAV remote sensing image based on DSP platform
    SUN Peng, XIAO Jing, ZHAO Haimeng, LIU Fan, YAN Lei, ZHAO Hongying
    2020, 40(4):  1237-1242.  DOI: 10.11772/j.issn.1001-9081.2019091689
    Asbtract ( )   PDF (1347KB) ( )  
    References | Related Articles | Metrics
    To satisfy the requirement of real-time and rapid processing of Scale-Invariant Feature Transform(SIFT) algorithm for remote sensing images of large-scale Unmanned Aerial Vehicle (UAV) network on the scene, an implementation scheme of the algorithm by using the hardware multiplier of Digital Signal Processor(DSP)kernel was proposed to process the multiplication of single-precision floating-point pixel data. Firstly,according to the characteristics of data input and output of the hardware multiplier with DSP kernel,the image data structure and the image function of SIFT algorithm were reconstructed in order to make the hardware multiplier perform the multiplication calculation of single-precision floating-point pixel data of SIFT algorithm. Secondly,the software pipelining technology was adopted to rearrange the iterative computation,so as to enhance the parallel computing ability of the algorithm. Finally,the dynamic data produced in the algorithm calculation process were transferred to the Double Data Rate 3 synchronous dynamic random access memory(DDR3)to enlarge the storage space of the algorithm data. Experimental results show that the SIFT algorithm on DSP platform is able to achieve high-precision and fast processing for 1 000×750 remote sensing images of UAV,and the scheme satisfies the requirement of real-time and rapid processing of SIFT algorithm for remote sensing images of UAV network on the scene.
2023 Vol.43 No.5

Current Issue
Honorary Editor-in-Chief: ZHANG Jingzhong
Editor-in-Chief: XU Zongben
Associate Editor: SHEN Hengtao XIA Zhaohui
Domestic Post Distribution Code: 62-110
Foreign Distribution Code: M4616
No. 9, 4th Section of South Renmin Road, Chengdu 610041, China
Tel: 028-85224283-803
Website: www.joca.cn
E-mail: bjb@joca.cn
Join CCF