Computer and Modernization

Blind Restoration of Defocused Target Images Based on Self-adaptive Blur Map Estimation

YANG Miao, YU Tao, ZHAO Li-qiang, GUO Yong-qi, QIU Ke-peng

2019, 0(11): 1. doi:10.3969/j.issn.1006-2475.2019.11.001

Asbtract ( 122 )

PDF (5299KB) ( 108 )

References | Related Articles | Metrics

In order to solve the problem of global and local defocused blur of target image, a fast blind restoration method based on self-adaptive blur map estimation is proposed. Firstly, according to the continuity of image edges in the scale space, the re-blur amount matrix is chosen self-adaptively, and the defocused blur target image is re-blurred. Then, the sparse blur map is calculated by the difference ratio in edges between blur and re-blur images, and the blur map is obtained by guided filtering. Finally, the physical relationship between the blur map and defocused target image is established based on the optics focal model, and the defocused blur image is restored quickly. The experimental results show that the proposed method can effectively restore the defocused blur target image and enhance the edge features of the target image, which has great advantages in algorithm operation efficiency and avoids the high time consumption of the iterative algorithm, and is suitable for practical industrial applications.

Non-reference Image Quality Assessment Models Based on Multi-task

YANG Lu, WEI Min

2019, 0(11): 7. doi:10.3969/j.issn.1006-2475.2019.11.002

Asbtract ( 186 )

PDF (2602KB) ( 130 )

References | Related Articles | Metrics

The Non-Reference Image Quality Assessment (NRIQA) models based on Deep Learning(DL) include two common structures, namely the single-task structure and the multi-task structure. In order to study the effect of the multi-task structure on model accuracy without pre-training, the performance of the MEON-adjusted multi-task model and the single-task model in the NRIQA task is compared and analyzed. The accuracy rate of the multi-task model reaches 0.882 and 0.871 respectively on LIVE and TID2013 image quality assessment databases. Experiments show that the multi-task model without pre-training still exhibits equal even better performance than the single-task model. On this basis, the sub-task output dimension experiment of the multi-task model shows that in NRIQA research, the sub-task can be pre-trained on the relevant datasets according to the demand and goal, and then fine-tuned in combination with the quality evaluation task, it has the advantage of being able to integrate into other tasks by transferring learning.

Cloud Removal Algorithm of Remote Sensing Image Based on GANs

LI Hua-ying1, LIN Dao-yu2, ZHANG Jie1, LIU Bi-xin1

2019, 0(11): 13. doi:10.3969/j.issn.1006-2475.2019.11.003

Asbtract ( 359 )

PDF (3980KB) ( 187 )

References | Related Articles | Metrics

Many problems in computer vision can be abstracted as “converting” an input image into a corresponding output image, which is a general solution to many computer vision problems, such as semantic segmentation, image style transfer, etc. In this paper, the remote sensing image cloud removal is used as the special case of image conversion, and the image conversion algorithm based on Generative Adversarial Networks (GANs) is studied. The GANs based on the residual module is proposed to remove the thick cloud and thin cloud from single remote sensing image. At the same time, the proposed multi-scale discriminator and VGG loss function can effectively deal with the cloud occlusion problem of complex scenes. The experimental results show that the proposed image conversion algorithm increases the peak signal-to-noise ratio on the remote sensing image thin cloud dataset by 1.64 dB and increases the peak signal-to-noise ratio by 1.92 dB on the thick cloud dataset. At the same time, the generated cloud-free remote sensing images have high structural similarity with the real cloud-free images.

Multi-source Data Integration Method Based on Data Virtualization Technology

ZHANG Zi-ye1, LIU Yu-long1, HU Bei2

2019, 0(11): 18. doi:10.3969/j.issn.1006-2475.2019.11.004

Asbtract ( 237 )

PDF (1051KB) ( 197 )

References | Related Articles | Metrics

There is no uniform format standard for judicial business data storage, and there exists data islands in each organization’s data query and access. In order to solve the heterogeneity between data access, this paper proposes a multi-source judicial data integration method based on data virtualization, which establishes metadata mapping relationship through data virtualization technology, and uses middleware to form a data exchange center to realize multi-organ and multi-type judicial data integration. The improved K-means clustering algorithm is used to cluster virtual object metadata, shorten data access time and improve judicial data query efficiency. The proposed method can ignore the influence of data storage heterogeneity and realize accessible data access channels of various judicial organs.

Combined Hydrological Time Series Forecasting Model Based on CNN and MC

XU Guo-yan1, ZHU Jin1, SI Cun-you2, HU Wen-bin2, LIU Fan1

2019, 0(11): 23. doi:10.3969/j.issn.1006-2475.2019.11.005

Asbtract ( 300 )

PDF (1098KB) ( 239 )

References | Related Articles | Metrics

Accurate forecast of water level is an effective measure to prevent flood disasters. Under the background of the continuous development of in-depth learning, a combined hydrological time series forecast model based on convolutional neural network and Markov chain is proposed. The model solves the problems that the existing algorithms do not consider the spatial correlation between stations, multi-dimensional input will increase the complexity of data reconstruction in feature extraction, and the single model only considers the linear part of water level time series without considering the non-linear part, which leads to the low forecast accuracy. Firstly, the combined model uses convolutional neural network to train water level time series and rainfall time series to predict future water level and calculates residual series with original time series. Then, the residual forecast results obtained by Markov chain training residual series and the value of convolution neural network forecast are added together to get the final result. Experiments show that this method can achieve better forecast accuracy than the existing algorithms.

Analysis and Application of Time Series Prediction Based on #br# Holt-Winters in Big Data Monitoring System

WANG Yu-fei1, DU Tian-cang1,2

2019, 0(11): 29. doi:10.3969/j.issn.1006-2475.2019.11.006

Asbtract ( 205 )

PDF (1164KB) ( 136 )

References | Related Articles | Metrics

According to the requirement of time series prediction accuracy and real-time and the trend and seasonal variation of time series in big data monitoring system, Holt-Winters algorithm is selected to build time series prediction model. First, this paper introduces the concept and characteristics of time series, then analyzes the principle of Holt-Winters algorithm and prediction conditions. Choosing the appropriate smoothing coefficient is the key to affect the accuracy of Holt-Winters algorithm. This paper introduces an algorithm for solving dynamic cubic smoothing coefficient in different time intervals by combining L-BFGS algorithm. Finally, the record of two-day page visits of users are used as experimental data. Through the analysis of relative error index, it is verified that the algorithm meets the requirements of big data monitoring system for time series prediction, and has better prediction application effect.

Improvement for Feature Point Extraction Based on Kinect 3D Reconstruction

CHEN Kai-yang1, LUO Zhi-zao2, WANG Jian-xing2

2019, 0(11): 34. doi:10.3969/j.issn.1006-2475.2019.11.007

Asbtract ( 176 )

PDF (668KB) ( 108 )

References | Related Articles | Metrics

To solve the problem of low performance-price ratio due to low efficiency and high distortion of feature points extraction in the complex indoor environment of robots, an improved SIFT feature points extraction and matching algorithm is proposed, and on this basis, a SLAM system based on Kinect is built.
The SLAM system front end improves the SIFT feature point extraction method, uses the Gaussian separation fuzzy function, improves the speed of SIFT algorithm to extract the feature point, and uses RANSAC to screen unstable feature points. The SLAM system with improved SIFT feature points extraction method can reconstruct the complex and empty indoor environment with high efficiency and low distortion.

A Graph-based Collaborative Filtering Algorithm in Movie Recommendation System

ZHENG Ce1,2, YOU Jia-li1,2

2019, 0(11): 38. doi:10.3969/j.issn.1006-2475.2019.11.008

Asbtract ( 184 )

PDF (1411KB) ( 129 )

References | Related Articles | Metrics

In order to solve the problem of sparse scoring data in the field of video service, the traditional collaborative filtering algorithm is usually used, but the video similarity calculation of the algorithm only uses score matrix, which results in low recommendution accuracy. In this paper, a graph based collaborative filtering algorithm is proposed for the scene of the movie in video resources. Combining the correlation between movie attributes and user preferences, the map elements of film information such as types, directors and actors are mapped, and the similarity of film resources is calculated by combining the features of graph structure. This method replaces the similarity calculation method of scoring matrix in traditional collaborative filtering algorithm, which alleviates the problem that the recommendation accuracy is affected by the sparse data. Experiment verifies the effectiveness of the proposed algorithm.

Combinational Software Reliability Models Based on Extreme Learning Machine

LI Si-yu

2019, 0(11): 44. doi:10.3969/j.issn.1006-2475.2019.11.009

Asbtract ( 160 )

PDF (714KB) ( 124 )

References | Related Articles | Metrics

To solve the problem of weak adaptability of single software reliability models and poor stability of data-driven models, this paper chooses three typical software reliability models as basic models, uses extreme learning machine to weigh and optimize the prediction results of basic models to obtain the combined software reliability model, which realizes the organic combination of classical software reliability models and artificial intelligence algorithm. Through simulation experiments on three sets of software failure data and comparison with the prediction results of single models, combinational models based on other neural network algorithms and data-driven model, it is verified that the combined software reliability model in this paper can effectively improve the prediction accuracy and model adaptability.

Software Adaptation Method Based on Control Theory

JIN Zhi-qiang, LI Xiao-hui, LYU Ren-jian, WANG Han, CHENG Kai

2019, 0(11): 49. doi:10.3969/j.issn.1006-2475.2019.11.010

Asbtract ( 164 )

PDF (1323KB) ( 107 )

References | Related Articles | Metrics

Software adaptation is a hot research topic in software field in recent years, but with the change of computing mode brought by edge computing, the current software adaptation methods are difficult to adapt to the equipment resources with huge differences. In order to improve the adaptability of software to equipment resources, the resource consumption of software is modeled and a resource adaptive method based on control theory is proposed, and a resource adaptive controller is designed according to PID control theory. Experiments show that the proposed software adaptive method and resource adaptive controller can adapt to the changes of resources in embedded environment under different load conditions, load changes and different equipment resources. Compared with existing software adaptive methods, the proposed method has better resource adaptive ability.

Detecting Method for Workpiece Size and Symmetry of Blower Rotor

ZHOU Hong-yu1, MA Qian-li1, MA Peng-bo2

2019, 0(11): 55. doi:10.3969/j.issn.1006-2475.2019.11.011

Asbtract ( 181 )

PDF (1580KB) ( 119 )

References | Related Articles | Metrics

In view of the fact that the size and symmetry of the blower rotor workpiece often appear to be inconsistent with the standard in industrial production, this paper uses the rotor workpiece images to design a rotor size and symmetry detection method based on the edge detection technology. In order to reduce the noise point of original image, the method performs noise reduction processing and binarization processing on the original image based on Gaussian filtering. On this basis, the edge detection algorithm based on Canny operator and the round fitting method based on the least square method are used to detect the holes inside the workpiece. At the same time, the improved Canny algorithm is used to detect the outer edge of the workpiece to obtain relatively continuous edge pixel points, and the comparison method and dynamic simulation experiment are designed to test the size and symmetry. The experiments show that the detection method proposed in this paper has a good effect on the detection of the size and symmetry of the rotor workpiece.

CP-ABE Scheme with Attribute Revocation Under Environment of Multi-attribute Authority

ZHOU Li-jing, WANG Min, QIN Lu-lu

2019, 0(11): 60. doi:10.3969/j.issn.1006-2475.2019.11.012

Asbtract ( 219 )

PDF (616KB) ( 101 )

References | Related Articles | Metrics

Attribute-Based Encryption (ABE) can not only guarantee the security of data, but also realize fine-grained access control of data. In reality, due to user attributes may be frequently changed, it is very important to implement attribute revocation in ABE scheme. To solve the problem how to realize effective user decryption and key escrow in the existing schemes in the devices with limited computational efficiency, this paper proposes a revocable ABE scheme under multi-attribute authorization in cloud environment. In the proposed scheme, the client uses the outsourcing decryption technology to reduce the local computing load, and entrusts the updating of the combination key and ciphertext to the cloud server to realize the function of attribute revocation. The security analysis shows that this scheme has indistinguishable security under chosen plaintext attack. The performance analysis shows that the proposed scheme is more efficient.

Machine Vision Application Framework of Desktop Robot Arm

MA Xing-lu， HE Ai-xin， LI Ying-ying

2019, 0(11): 64. doi:10.3969/j.issn.1006-2475.2019.11.013

Asbtract ( 186 )

PDF (1154KB) ( 107 )

References | Related Articles | Metrics

Compared with the traditional industry robot arm, the desktop robot arm has the characteristics of changeable environment and man-machine cooperation, which makes it more important to increase its visual function. At present, there are many application frameworks to realize machine vision. How to build an appropriate software and hardware platform for visual application according to the working environment and nature of the robot arm, so as to improve the accuracy and efficiency of machine vision recognition is the focus of this paper. Based on TensorFlow’s deep learning framework, the embedded system’s software and hardware design, and the application of OpenCV and other image processing software, this paper constructs a machine vision secondary development framework suitable for desktop robot arm, which provides a foundation for the further development of vision-based robot arm applications. Through simulation test and case application of man-machine cooperation, it is shown that the framework has better adaptability and better efficiency.

A Pedestrian Detection Method in Intelligent Video Monitoring System

YANG Lei1, WANG Shao-yun2, LIU Li-ran1, GONG Yong-fu1

2019, 0(11): 69. doi:10.3969/j.issn.1006-2475.2019.11.014

Asbtract ( 231 )

PDF (2636KB) ( 171 )

References | Related Articles | Metrics

In the pedestrian detection of intelligent video monitoring system, the current target detection algorithms R-CNN and YOLO series are slower, which cannot meet the real-time requirements, or require large GPU memory space, which is difficult to deploy. YOLOv3-tiny algorithm, as a simplified version of YOLO series, has less requirements for equipment and is faster, but its accuracy is low. In this paper, the number of horizontal and vertical directions of YOLOv3-tiny algorithm grid cell is adjusted, the network structure of YOLOv3-tiny algorithm is optimized, and the number and size of anchors are determined by clustering, so as to obtain the improved YOLO-Y algorithm, and expand the training data set by data enhancement method. The improved YOLO-Y algorithm improves the mAP from 90% to 92%, Recall from 95% to 97%, detection speed up to 26 frames/s, occupies about 1 GB of video memory space. The experimental results show that the improved YOLO-Y algorithm significantly improves the detection accuracy of the algorithm, has real-time performance, and does not need too much memory space to meet the requirements of most intelligent video monitoring systems.

Dynamic Gesture Recognition Based on 3D Convolutional Neural Networks

GU Chen-nan, ZENG Xiao-qin

2019, 0(11): 75. doi:10.3969/j.issn.1006-2475.2019.11.015

Asbtract ( 275 )

PDF (1698KB) ( 266 )

References | Related Articles | Metrics

In video recognition, the traditional 2D convolution neural networks are easy to lose the relevant feature information in time dimension, which leads to the reduction of recognition accuracy. This paper uses 3D convolutional neural network as a basic network framework with 3D convolution kernel to extract the temporal and spatial features of videos, at the same time, the integration of multiple 3D convolutional neural network models are proposed to recognize dynamic gesture. In order to improve the convergence speed of the model and the stability of training, the network is optimized by Batch Normalization (BN) technology to shorten the training time of the network. Experimental results show that the proposed method has a good recognition performance for dynamic gesture recognition, and the recognition accuracy reaches 98.06% in Sheffield Kinect Gesture (SKIG) data set. Solely compared with RGB information, depth information and traditional 2D CNN, the gesture recognition rate is higher, which verifies the feasibility and effectiveness of the proposed method.

Intelligent Dynamic Estimation of Traffic Light Time Based on TAN Classifier

CHEN Hai-yang1, HUAN Xiao-min1, CHEN Xin-zhan2, LIU Xi-qing1

2019, 0(11): 81. doi:10.3969/j.issn.1006-2475.2019.11.016

Asbtract ( 122 )

PDF (1228KB) ( 95 )

References | Related Articles | Metrics

Aiming at the problem that traffic lights are weak in intelligence and easy to cause traffic jam, a time intelligent dynamic estimation method for traffic lights based on TAN classifier is proposed. Firstly, this paper analyzes the main factors which affect the traffic light time, and discretizes the collected data by fuzzy classification function. Then, it learns the structure of the TAN classifier by K2 algorithm. Next, it learns the parameters of TAN classifier with maximum likelihood estimation. Finally, the best traffic light time is estimated online by forwards-backwards algorithm based on the sliding window. The experimental simulation shows that the proposed method can dynamically estimate the optimal traffic light time according to real-time traffic information. When the traffic light is free, the traffic light time is short, otherwise, time is long. It is effective to alleviate traffic jam and reduce environment pollution.

Digitalized Design of Solar Terms Culture Genes

QU Li-na, PENG li

2019, 0(11): 88. doi:10.3969/j.issn.1006-2475.2019.11.017

Asbtract ( 208 )

PDF (4807KB) ( 85 )

References | Related Articles | Metrics

The paper is to integrate the cultural gene of 24 solar terms into the digital design, improve the participation and positivity of users, expand the scope and speed up the speed of its transmission, so as to achieve better protection and inheritance of the 24 solar terms culture. Through perceptual analysis and shape grammar, after deep analysis, we extract the connotation, color and morphological genes of the 24 solar terms and design solar term culture gene in combination with digital technology. The existing solar digital products are scored by the Likert five-point scale method. By comparing and analyzing each measurement index, the solar energy design example “Jie Qi Baby” is evaluated. This idea not only saves the cost of intangible cultural heritage protection, but also achieves the best protection effect, which provides a new reference for the protection of other intangible cultural heritage.

A Substation Simulation Training System Based on HoloLens

LEI Zhen-jiang1, WANG Lei1, CUI Ji-sheng1, CHEN Shuo2, XU Zheng-qing3, YU Hai-long2, MENG Hao2, LIU Guo-zhong2

2019, 0(11): 94. doi:10.3969/j.issn.1006-2475.2019.11.018

Asbtract ( 239 )

PDF (2305KB) ( 136 )

References | Related Articles | Metrics

Microsoft HoloLens is an augmented reality PC headset that brings high-definition 3D digital holograms to life in the real world, therefore, it is adopted as the platform to develop the substation operation training system based on the augmented reality technology. The main camera of HoloLens is used as the image acquisition equipment of the substation, and the identification methods of the local feature of equipment based on ASIFT, the artificial marker and the equipment label based on OCR are combined to realize the reliable identification of the substation equipment with different size and shape. The object equipment in the substation is accurately located and registered initially by the feature point recognition of artificial marker or equipment based on vision, and the Anchor technology of HoloLens is used to realize fast tracking and registration of target devices. The learning, training and assessment environment of substation based on augmented reality is constructed, and demonstration and verification are carried out. By expanding the functions of the existing training system, this paper improves the practicability, interactivity and experience of the existing training system, and improves the training effect and efficiency.

Effectiveness of Interactive Environmental Learning Based on #br# Visual Auditory Sensory Channels

YANG Qin1, WANG Si-yu2, XIONG Wen-bin2

2019, 0(11): 100. doi:10.3969/j.issn.1006-2475.2019.11.019

Asbtract ( 149 )

PDF (987KB) ( 74 )

References | Related Articles | Metrics

This paper studies the impact of interactive environment on learning outcomes, and explores the effectiveness of learning outcomes in non-interactive teaching, interactive teaching and teaching environment with different levels of interaction through the channel effect produced by change of tester’s auditory and visual sensory combination. This paper designs a combined teaching test, analyzes the learning effect of learners by combining the combination of different visual and auditory channels with the teaching methods with different levels of interactivity. The experiment in this paper shows that the channel effect can promote learning in the interactive learning environment, and the learning environment with different levels of interaction can also have different effects on the channel effect. The interactive learning environment with a certain degree would promote learning, while the learning environment with higher interaction degree would have negative effects.

Construction of Real Estate Data Sharing System Based on Internet and E-government

XU Han1,2, LIU Cong-jun 1,2

2019, 0(11): 106. doi:10.3969/j.issn.1006-2475.2019.11.020

Asbtract ( 196 )

PDF (1790KB) ( 111 )

References | Related Articles | Metrics

With the development of the “Internet〖KG-*3〗+government service” mode and the in-depth reform of information technology, the real estate registration institution should establish an interoperability mechanism among various departments to solve the difficult problem of information sharing among various departments in the same area of real estate institutions. By establishing an integrated information management platform for data sharing, this paper connects the application systems of various departments with the data sharing system through data interface and other technical means, ensures the unification of business and data, realizes the exchange, sharing and orderly connection of information and data among systems, and promotes the comprehensive, coordinated and sustainable development of government service informatization.

Packet Drift Strategy Based on Packet Cluster Mapping Architecture

LIANG Ci1, CHEN Shi-ping2

2019, 0(11): 112. doi:10.3969/j.issn.1006-2475.2019.11.021

Asbtract ( 128 )

PDF (1584KB) ( 124 )

References | Related Articles | Metrics

Resource allocation in data centers has always been a hot issue at home and abroad. In view of this, a package drift strategy is proposed in the framework of packet cluster. Firstly, this strategy degrades the mapping complexity between virtual machines and servers by using the hierarchical idea of packet cluster. According to the resource load on the cluster, FCM method is used to partition the cluster. Then, according to the maximum correlation algorithm, packets are selected to join the queue to be drifted, and the processing priority of the queue is set. Finally, a probability model is constructed to select the best target cluster for the drift package according to the drift cost and resource matching degree. Experiments of the packet drift strategy on CloudSim simulation platform show that the proposed scheme can effectively improve the service quality and resource utilization of data centers, and also has good performance in reducing energy consumption.

Construction and Application of Experimental Platform #br# Based on Packet Cluster Mapping Mechanism

DING Shun1, CHEN Shi-ping2

2019, 0(11): 120. doi:10.3969/j.issn.1006-2475.2019.11.022

Asbtract ( 150 )

PDF (1641KB) ( 114 )

References | Related Articles | Metrics

Starting from the actual requirements of general cloud platform and project, combining with the packet cluster architecture, and using CloudStack open source cloud platform, an experimental platform based on packet cluster mapping mechanism is designed. The experimental platform adopts a layered design method, including hardware facilities layer, virtual resource layer, scheduling layer, middle layer of packet cluster, and user application layer, which converts the traditional application for resources in the form of virtual machine into the application in the form of demand packet, and the user can specify the structure of required packages and the physical resources required for each package. Through the analysis of the scheduling principle of the cloud platform, it is expounded how to apply the cluster deployment algorithm involved in the project to the experimental platform, which provides experimental basis for the research results of the subsequent improvement project. Finally, the six important management functions of the cloud computing management platform are selected. Through the overall functional test of experimental platform based on the clustering mechanism, CloudStack and OpenStack cloud management platforms, the results show that the clustered experimental platform in this paper provides comprehensive management functions and has certain application market.

Table of Content