Loading...

Table of Content

    22 April 2020, Volume 0 Issue 04
    A Squint Object Robust Detection Method Based on #br# Perspective Transformation Data Augmentation
    LI Cheng-qi1, ZHENG Wen-jie1, HUANG Wen-li2, WEN Zhao-yang2
    2020, 0(04):  1.  doi:10.3969/j.issn.1006-2475.2020.04.001
    Asbtract ( 142 )   PDF (1987KB) ( 263 )  
    References | Related Articles | Metrics
    Object detection makes great progress in the accuracy of recognition by using convolutional neural network technology. The general object detection has achieved better detection results, but the algorithm detection effect is poor for the strabismus object problem with less sample size in industrial production. The main reason is that the training samples are very rare, resulting in shift of the detection model training based on deep neural network, which affects the overall detection accuracy. This paper proposes a squint object robust detection method based on perspective transformation data augmentation. It can solve the problem of less strabismus object sample size by perspective transformation to simulate the scene of strabismus object, increase the squint object sample size for training, and improve the accuracy of recognition of squint objects. Experiments show that the proposed method has obvious improvement effect on detection accuracy.
    Speech Tracking Based on Cluster Analysis and Speaker Recognition
    HAO Min, LIU Hang, LI Yang, JIAN Dan, WANG Jun-ying
    2020, 0(04):  7.  doi:10.3969/j.issn.1006-2475.2020.04.002
    Asbtract ( 242 )   PDF (1216KB) ( 180 )  
    References | Related Articles | Metrics
    At present, the speech tracking quality will be seriously reduced under the condition of speaker interference, that is, mixed speech signals of multiple speakers in a speech segment. Aiming at this situation, a speech tracking algorithm based on cluster analysis and speaker recognition is proposed. Firstly, the improved clustering analysis method is used for speech separation. Specifically, it includes caching the center of mass and lowering the sampling rate in K-means clustering, and introducing regular terms into embedding feature space. Secondly, the GMM-UBM speaker model is used for speech tracking. The experimental results show that the improved cluster analysis method can effectively improve the real-time performance of the algorithm and the quality of speech separation, the GMM-UBM model has an 84% recognition rate in 3 s speech test.
    A Distributed Sustainable Integrated Automated Testing Platform
    LEI Jian-sheng1, SU Xiao2, JIN Ming-lei1
    2020, 0(04):  14.  doi:10.3969/j.issn.1006-2475.2020.04.003
    Asbtract ( 292 )   PDF (1092KB) ( 139 )  
    References | Related Articles | Metrics
    In the process of software development, with the increasing logical complexity, special testing can effectively reduce software bugs and improve the quality of software. However, the traditional manual testing can not meet the needs of current software development. Therefore, a distributed sustainable integration automation testing platform is proposed. Based on Jenkins platform, the distributed sustainable integration is realized. Git is selected as a version management tool and Katalon is used as an automation testing tool. The platform realizes daily automatic updating and continuous integration automation testing of Web software. The platform greatly reduces the workload of the testers to repeat the regression test in Web software testing.
    A Real-time Video Stitching System in Large Parallax Scene
    YANG Dan1,2, CHEN Jun1,2, ZHU Xiao-yong1
    2020, 0(04):  19.  doi:10.3969/j.issn.1006-2475.2020.04.004
    Asbtract ( 287 )   PDF (2211KB) ( 159 )  
    References | Related Articles | Metrics
    With the development of virtual reality technology and live video technology, panoramic video live broadcast has received extensive attention. Traditional panoramic products are prone to image distortion when dealing with large parallax scenes, and they are difficult to balance real-time requirements. In order to solve these problems, a two-way video splicing system is designed. Firstly, a mosaic background model is established. Then, the image registration is completed by combining the ORB (Oriented FAST and Rotated BRIEF) with APAP (As-Projective-As-Possible) based on moving DLT (Direct Linear Transformation) algorithm. A method based on minimum energy detection is improved to find the optimal seam to avoid ghosting and misalignment caused by moving foreground. Finally, by splicing the parameter index table obtained in the stage of the model calculation, the overlapping regions are fused to complete the real-time splicing of video frames. The experimental results show that the proposed system can handle large parallax scenes and has good real-time stitching effect.
    An Image Style Conversion Technology Based on EBGAN
    TAO Ying, LIU Hui-yi
    2020, 0(04):  24.  doi:10.3969/j.issn.1006-2475.2020.04.005
    Asbtract ( 219 )   PDF (6679KB) ( 143 )  
    References | Related Articles | Metrics
    In order to solve the problem of poor diversity of the generated images in the traditional image style conversion algorithm, this paper proposes a network model based on EBGAN (Energy-Based Generative Adversarial Net). The idea of energy function is introduced into the discriminator, and the Autoencoder is designed to generate different reconstruction results for the true and false input respectively. The error value before and after the reconstruction of the input image is calculated, which is used as the energy concept to identify the input image. In the coding stage of Autoencoder, the orthogonal control is introduced in to the encoded vectors to control the maximum orthogonalization of two vectors in the same batch, so as to promote the generator net to generate images in different directions. Experiments on Facades and Cityscapes datasets show that the proposed network model can effectively achieve process of image stylization and generate more diversified images than the traditional network model.
    Face Recognition and Tracking System of Photographic Robot #br# Based on YOLOv3 and ResNet50
    CHEN Kai, ZU Li, OU Yi
    2020, 0(04):  30.  doi:10.3969/j.issn.1006-2475.2020.04.006
    Asbtract ( 421 )   PDF (2453KB) ( 166 )  
    References | Related Articles | Metrics
    In the virtual studio,aiming at the task that the photographic robot needs completing automatical face recognition and shot tracking of the host, a system of face recognition and shot tracking for the host based on YOLOv3 face detection and ResNet50 construction is proposed. In order to improve the accuracy of face recognition for photographic robots on open sets, a host face training set based on CASIA-FaceV5 and PubFig data sets is constructed, and the model is trained on modified ResNet50 with joint supervision. An experiment is carried out by combining with the motion control algorithm of photographic robot, the experiment shows that the face recognition tracking system has robust recognition accuracy and real-time performance, and can meet the requirements of face tracking of photographic robot in the virtual studio.
    An Improved LSTM Model in the Application of Image Caption Generation
    WANG Zhi-ping, ZHENG Bao-you, LIU Yi-wei
    2020, 0(04):  37.  doi:10.3969/j.issn.1006-2475.2020.04.007
    Asbtract ( 265 )   PDF (918KB) ( 192 )  
    References | Related Articles | Metrics
    In order to solve the problem of premature saturation of traditional Long Short-Term Memory(LSTM) neural network and generate a more accurate description for a given picture, this paper proposes a long short-term memory neural network model based on inverse tangent function(ITLSTM). Firstly, the classical convolutional neural network model is used to extract image features. Then, the ITLSTM neural network model is used to characterize the corresponding description of the image. Finally, the performance of the model is evaluated on the Flickr8K dataset and compared with several classic image caption generation models such as Google NIC. The experimental results show that the proposed model can effectively improve the accuracy of image caption generation.
    A Fuzzy Keyword Search Encryption Scheme Without Secure Channel
    CAO Yong-ming
    2020, 0(04):  42.  doi:10.3969/j.issn.1006-2475.2020.04.008
    Asbtract ( 232 )   PDF (530KB) ( 145 )  
    References | Related Articles | Metrics
    Today with highly developed Internet technology, people are more and more accustomed to upload personal data to the cloud for storage. The traditional searchable public key encryption scheme only supports searching for exact keywords. Users need to input search keywords accurately and use secure channels to transmit traps, which reduces the availability of the system. For secure channel, combining with public key encryption technology, this paper propose a scheme of searching and encrypting fuzzy keywords without secure channel. This scheme can guarantee the privacy of information without using secure channel, and will use wildcard technology to reduce the space size of keyword set. The security verification is also given.
    Identifier Resolution Model Based on Encryption Transmission
    ZUO Peng, HE Zhi-mou, YUAN Meng, ZHANG Hai-kuo, YANG Wei-ping
    2020, 0(04):  46.  doi:10.3969/j.issn.1006-2475.2020.04.009
    Asbtract ( 217 )   PDF (1060KB) ( 140 )  
    References | Related Articles | Metrics
    Identifier resolution system is the foundation for the stable operation and development of Internet. The privacy protection and data security issues of identifier resolution system draw significantly more attention. Technologies including DNSSEC based on digital signature and DoT based on encryption technology can solve the security problem partially, but cannot realize the user privacy protection of the whole process of identifier resolution. Based on the current status of technology research, a new identifier resolution trust model based on encryption transmission is proposed, and a trust chain is established to realize the trust transfer of each node in the identifier resolution system, and through the whole process of encryption communication, the user privacy and data security during identifier parsing are protected. Firstly, the research status of security technology in domain name area is introduced, then the whole structure, trust chain model and work flow of the proposed model are described, finally by five group experiments, the delay, performance and security of the model under different encryption methods and transfer protocols are tested and analyzed, and the feasibility of the model is verified by combining with the test results of live DNS.
    Trusted Secure Access Scheme of Ubiquitous Power IoT
    WU Jin-yu1, ZHANG Li-juan2, SUN Hong-di2, LAI Yu-yang2
    2020, 0(04):  52.  doi:10.3969/j.issn.1006-2475.2020.04.010
    Asbtract ( 296 )   PDF (1226KB) ( 187 )  
    References | Related Articles | Metrics
    In order to comprehensively improve the comprehensive security defense capability of all-service ubiquitous power Internet of Things(IoT), and solve the lack and deficiency of the security protection guidance and terminal authentication mechanism of all-service ubiquitous power IoT, this  paper proposes a trusted secure access scheme of ubiquitous power IoT. Firstly, a unique fingerprint information is determined for the power IoT terminal layer equipment. Then, combined with the fingerprint information, the identification and password technology is used to realize the access authentication of terminal layer equipment and block the access of illegal terminals. Finally, the security transfer mechanism of legal terminal identity information is designed to trace the abnormal behavior of legal terminal according to the identity information.
    A Public Integrity Auditing Scheme Based on Blind Signature for Shared Data in Cloud
    ZHANG Xi, WANG Jian
    2020, 0(04):  60.  doi:10.3969/j.issn.1006-2475.2020.04.011
    Asbtract ( 185 )   PDF (1041KB) ( 118 )  
    References | Related Articles | Metrics
    Public integrity auditing for shared data in the cloud is used to verify the integrity of data which are shared among group users in the cloud. Traditionally, group users need to generate the authenticator for each data block, and then upload shared data and the corresponding authenticators to the cloud server for storage. However, the users computing resources are limited and computing power is not enough, so it takes a lot of computing overhead for users to generate authenticators for data blocks. In order to save users computing resources and improve the efficiency of authenticator generation, a public integrity auditing scheme based on blind signature for shared data in the cloud is proposed. Users firstly blind the data blocks and then send them to the authenticator generation center to generate the corresponding authenticators. In addition, the Third Party Auditor (TPA) is authorized to audit in this scheme, which effectively avoids the DDoS attacks of the attacker on the cloud sever. The security analysis and experimental results show that the proposed scheme is safe and efficient.
    A Keyword Extraction Algorithm Based on Adaptive Related Entropy
    LUO You-zhi 1,2, CHEN Zheng-ming2, CHEN Ming2, MEI Wen-tao2
    2020, 0(04):  67.  doi:10.3969/j.issn.1006-2475.2020.04.012
    Asbtract ( 224 )   PDF (782KB) ( 138 )  
    References | Related Articles | Metrics
    Compared with the traditional technique of keyword extraction based on vocabulary frequency size, the TexRank algorithm can consider the similarity information between vocabulary nodes, but ignores vocabulary context information and the semantic structure of the article. On the basis of the weighted iteration of node diagram, this paper uses the association rule information of text context vocabulary, introduces the concept of association entropy, adaptively adjusts damping coefficient and sliding window size. It is closer to the actual semantic situation of text vocabulary, and can better deal with low word frequency and new vocabulary information. Experimental result shows that compared with TFIDF and TR algorithm, this method can achieve more accurate results when processing keyword extraction.
    Network Traffic Prediction Based on SCSO-GRU Model
    GAO Bai-hong1, LIU Zhao-hui1, LIU Hua2
    2020, 0(04):  72.  doi:10.3969/j.issn.1006-2475.2020.04.013
    Asbtract ( 258 )   PDF (1051KB) ( 177 )  
    References | Related Articles | Metrics
    Network traffic has the characteristics of real-time, instability and correlation. The traditional prediction model of network traffic has the shortcomings of weak generalization ability and low prediction accuracy. To overcome these shortcomings, a network traffic prediction model (SCSO-GRU) based on GRU neural network combined with Sine-Cosine Swarm Optimization (SCSO) algorithm is proposed. Firstly, this paper introduces the particle update principle of SCSO algorithm. Then, it constructs a network traffic prediction model with SCSO-GRU neural network. The SCSO algorithm is used in model training to improve the training effect and overcome the disadvantage that the traditional GRU neural network converges to local optimum. Finally, this paper uses the SCSO-GRU model to predict the network traffic. The experimental results show that compared with the traditional LSTM and GRU models, the proposed model has better convergence efficiency and prediction accuracy, and can better describe the trend of network traffic.
    Risk Situation Assessment Method of Field Operation Based on Data Mining
    JIANG Yi, OU Yu-qiang, LIANG Guang, GAO Yang, YAN Yong-gao, LIN Jie, ZHAO Xiao-ning
    2020, 0(04):  78.  doi:10.3969/j.issn.1006-2475.2020.04.014
    Asbtract ( 223 )   PDF (1301KB) ( 156 )  
    References | Related Articles | Metrics
    In order to ensure the safe and stable operation of power grid and reduce production safety accidents, power supply companies carry out on-site safety management for a large number of infrastructure, production and operation operations, and have accumulated massive operational management and safety management business data. In order to make better use of these data and make it truly serve for early warning, prevention and control of operational risks of power supply enterprise, on the basis of combing the data of the entire process of field operations of power supply companies, the historical data of operational risk situation factors are analyzed to identify risk influencing factors such as human, machine, material, law, environment and management. The field operation risk assessment index system and operation safety situation index model are constructed from six dimensions including personnel, equipment (machinery), materials and appliances, operation methods, operation environment,  and time management. Combined with information technology, the tool design of intelligent risk situation assessment is carried out.
    A Water Change Detection Method of SAR Images Based on #br# New Difference Operator and Texture
    LI Ling-yu, ZHANG Yi
    2020, 0(04):  85.  doi:10.3969/j.issn.1006-2475.2020.04.015
    Asbtract ( 187 )   PDF (7105KB) ( 112 )  
    References | Related Articles | Metrics
     Aiming at the problems of isolated noise points, artificial selection of partial parameters and incomplete information utilization in multi-temporal Synthetic Aperture Radar (SAR) change detection, an improved water change detection method of SAR images based on new difference operator and texture is proposed. First, according to characteristics of SAR images, combining with Log Ratio (LR) operator and Logarithmic Likelihood Ratio (LLR) operator, a new difference operator is proposed to amplify the characteristics of unchanged and changed regions. Then, according to the ratio graph of the histogram at two adjacent gray level in the new difference image, the threshold of initial segmentation is determined. Second, a new Fuzzy Local Information C-Means (FLICM) clustering method is proposed. This method utilizes the threshold from the previous step to obtain the initial clustering center. Then the texture-based FLICM method (FLICM_texture) is proposed to divide the difference image into three categories. Third, this paper divides the transition region again according to the threshold obtained by the difference image. This paper utilizes the SAR images over Canadas Ottawa, the Switzerland’s Bern and Chennai to demonstrate the superiority of this method. The Percentage of Correct Classifications (PCC) of Ottawa is 98.00%, the kappa coefficient is 92.03%. In Bern, the PCC can reach 99.66% and the kappa coefficient is 85.77%. In Chennai, the PCC can reach 98.83% and the kappa coefficient is 84.96%.
    Semantic Feature Extraction Algorithm for 3D Human Body Based on Template Matching
    LI Ling-jie1, TONG Jing1,2, BU Wen-yu1, SUN Hai-zhou1, CHEN Zheng-ming1,2
    2020, 0(04):  95.  doi:10.3969/j.issn.1006-2475.2020.04.016
    Asbtract ( 293 )   PDF (3202KB) ( 186 )  
    References | Related Articles | Metrics
    With the development of society and economy and the improvement of living standards, people’s demand for customized products is increasing, such as advanced customization of clothing and customization of fitness programs. In order to better meet these needs and provide accurate data support, it is necessary to extract precise human body semantic features, that is, a series of parameters such as height, chest size, and waist size of the human body. The existing feature extraction algorithms are analyzed and researched, and a semantic feature extraction algorithm for 3D human body based on template matching is designed. The template model is used to approximate the input model, and the semantic feature sampling points on the template model are extended to the input model. The NURBS curve is used to fit the sampling points to calculate the curve length. The experimental results show that the proposed algorithm has good comprehensive performance and can provide accurate and extensive data support for clothing customization, human body animation, and ergonomic design.
    Defect Detection Method for Medical Plastic Bottle Manufacturing Based on ResNet Network
    FU Lei, REN De-jun, HU Yun-qi, GAO Ming, QIU Lyu
    2020, 0(04):  104.  doi:10.3969/j.issn.1006-2475.2020.04.017
    Asbtract ( 269 )   PDF (1338KB) ( 243 )  
    References | Related Articles | Metrics
    This paper proposes a recognition method based on deep learning for the real-time detection of production defects such as medical plastic bottle bubbles and accumulated materials, designs the visual inspection hardware platform of the industrial site, describes the principle of the accumulation and bubble detection algorithm, and briefly describes the image pre-processing before the algorithm detection. Under the Pytorch framework, the real-time performance of aggregate detection is compared by orthogonal experiment between ResNet series algorithms and MobilenetV2 algorithm, and the detection performance of RetinaNet network on the bubbles is optimized.At the production site, the average detection accuracy of the proposed method is 99.7% and the single detection time is 29.7 ms. The Fβ index of the bubble is 99.5% and the single detection time is 35.5 ms, which meets the requirements of enterprise production.
    Physical Layer Algorithm Based on Channel Feature and Random Interpolation
    WU You, CHEN Cheng, JIN Long
    2020, 0(04):  109.  doi:10.3969/j.issn.1006-2475.2020.04.018
    Asbtract ( 155 )   PDF (1020KB) ( 95 )  
    References | Related Articles | Metrics
    The article studies the present safety situation of wireless communication system and the necessity of protecting physical layer security. Combined with the secret key generation method of wireless channel, a physical layer security algorithm for random interpolation based on the parallel modulation mode of Orthogonal Frequency Division Multiplexing (OFDM) system is proposed. The core idea of the algorithm is as follows, under the control of public secret key, data symbols which are output after Inverse Fast Fourier Transform (IFFT) are interpolated randomly, and the original OFDM symbol is reconstructed to make it difficult for illegal users to correctly demodulate the signal, so as to protect the security of transmission information. The algorithm is based on physical layer encryption, which can better protect air interface and wireless link. And the process of parallel encryption can reduce the complexity of communication system implementation. The results of theoretical analysis and simulation show that the algorithm can effectively withstand all kinds of illegal attacks, has little influence on the original performance of communication system, can be well adapted to multipath channel, and has a good ability to withstand multipath fading.
    Loop Closure Detection Algorithm Based on Mixed Global Pooling
    SONG Zhou-rui
    2020, 0(04):  115.  doi:10.3969/j.issn.1006-2475.2020.04.019
    Asbtract ( 205 )   PDF (1018KB) ( 399 )  
    References | Related Articles | Metrics
    The deep learning based loop closure detection algorithm has been verified to be superior to traditional methods. However, the computation burden of deep learning is heavy, so it is often difficult to deploy large convolutional neural networks on mobile robots, while small convolutional neural networks perform poorly in large scenes. Therefore, this paper proposes a scheme to deploy large convolutional neural networks on mobile robots. Firstly, the feature graph is transformed into the feature vector by using the mixed global pooling layer. Experiments show that the performance of this method is equivalent to that of other more complex methods and the calculation is simpler. Then, a block-based floating-point convolutional neural network acceleration engine is proposed, which significantly reduces the computational energy consumption and causes almost no performance loss without retraining.
    Surface Defect Detection of Empty Bottles Based on Improved SSD Algorithm
    WU Hua-yun, REN De-jun, FU Lei, GAO Ming, LYU Yi-zhao, QIU Lyu
    2020, 0(04):  121.  doi:10.3969/j.issn.1006-2475.2020.04.020
    Asbtract ( 261 )   PDF (1273KB) ( 207 )  
    References | Related Articles | Metrics
    Injection empty bottles can produce a large number of defects on the surface of bottles during production, and these defects have a significant impact on the appearance and use of the product. The traditional manual detection is no longer applicable due to the shortcomings of high labor intensity and low detection efficiency. For the traditional scene detection algorithms based on machine vision, the extracted features are often difficult to be used for defect classification and recognition for complex scene changes. Therefore, this paper proposes an SSD-based algorithm to detect the surface defects of injection empty bottles. Considering that the surface defect of empty bottle is small, it is difficult to extract features. In order to improve the detection effect, a feature fusion module is added to the SSD network structure to provide rich semantic features for the prediction layer. At the same time, an attention mechanism is introduced in the network to increase the feature extraction capability of the network and effectively improve the detection accuracy of the network. The method of this paper is verified on the empty bottle surface defect data set. The accuracy rate is 98.3%, the missed detection rate is 0.74%, the false detection rate is 0.96%, and the mAP is 96.5%. Compared with the mAP of original SSD algorithm, the algorithm in this paper improves by nearly 5.6 percentage points.