Computer and Modernization

Object Detection of Remote Sensing Image Based on Dual Attention Mechanism

ZHOU Xing, CHEN Li-fu

2020, 0(08): 1-7. doi:10.3969/j.issn.1006-2475.2020.08.001

Asbtract ( 440 )

PDF (3870KB) ( 332 )

References | Related Articles | Metrics

Aiming at the problem of low accuracy of small target detection in remote sensing image under complex background, an new SSD detection algorithm based on dual attention mechanism model is proposed. The algorithm introduces the dual attention mechanism model in the front-end feature extraction network. It strengthens the effective feature information of small targets in the low-level feature map and suppresses the redundant semantic information to achieve adaptive feature learning. In addition, dilated convolution is introduced into the spatial attention model to ensure the sensitivity of convolution kernel and reduce the parameters of the network. The Focal loss function is introduced as the classification loss function of the improved algorithm to improve the imbalance of samples and increase the weight ratio of positive and difficult samples during training, it promotes the detection performance of the algorithms. The detection results of the remote sensing image data set NWPU VHR-10 show that the improved algorithm not only ensures the detection speed, but also improves the detection accuracy. Compared with the traditional SSD algorithm, the mAP of the improved SSD algorithm is increased by 2.25 percentage points to 79.65%.

Abnormal Object Detection Method for Transmission Line

LI Hui, ZHOU Hang, DONG Yan, ZHANG Shu-jun

2020, 0(08): 8-13. doi:10.3969/j.issn.1006-2475.2020.08.002

Asbtract ( 425 )

PDF (4774KB) ( 276 )

References | Related Articles | Metrics

Abnormal object detection of transmission lines is an important part of power system monitoring. However, the existing detection methods are not designed effectively for transmission line scenes. There are some problems such as insufficient features extracted by the depth network and lack of robustness under the influence of variable target environment and scale changes. This paper proposes an abnormal object detection method for electric transmission line, using HRNet as the backbone network to extract high-resolution features, combined with HRFPN to optimizing the quality of object feature representation, and balancing the proportion of positive and negative anchor count generated in the RPN, using cascaded object detectors for classification and bounding box regression. The test results in the transmission line scene show that the proposed detection method has higher detection performance, which performs better than Faster R-CNN and Cascade R-CNN.

Simulation Method of Target SAR Image Based on Spectral Normalization Generative Adversarial Network

SUN Zhi-bo, XU Xiang-hui,

2020, 0(08): 14-20. doi:10.3969/j.issn.1006-2475.2020.08.003

Asbtract ( 369 )

PDF (3159KB) ( 249 )

References | Related Articles | Metrics

In order to solve the data sparse problem in Synthetic Aperture Radar (SAR) Automatic Target Recognition (ATR), this paper proposes a simulation method of target SAR images based on SN-GAN (Spectral Normalization Generative Adversarial Network). The method obtains the scattering intensity distribution maps by constructing the coupled physical model among target, scene and radar, then refines the scattering intensity distribution maps by using SN-GAN to generate the high-quality simulated SAR images. The similarity evaluation of the simulated images is carried out by 3 kinds of similarity evaluation algorithms to verify the effectiveness of the simulation method. Finally, through multiple sets of SAR ATR experiments, it is verified that adding simulated SAR images optimized by SN-GAN to the training set can effectively alleviate the data sparse problem and improve the accuracy of the classification algorithms.

3D Visualization of High Precision Map Lanes

ZHANG Cheng, MIAO Xin-rui, MOU Xue-man

2020, 0(08): 21-25. doi:10.3969/j.issn.1006-2475.2020.08.004

Asbtract ( 418 )

PDF (2982KB) ( 224 )

References | Related Articles | Metrics

In order to realize the visualization of 3D maps in the development of autonomous vehicles and simulate real roads in the simulator, a set of data restoration techniques is proposed. By transforming the latitude and longitude (WGS84) coordinate system of the original data into the ECEF coordinates, the 3D modeling software (Blender) API is used to automatically generate the polygon mesh without compiling the ESRI data. At the same time, in order to improve the efficiency of data use, further NEU coordinate segmentation is performed to maximize the convenience of the mesh. The generated 3D mesh is used respectively by the physics modules and rendering modules of the simulator. For the requirements of high visual quality, a method of interpolation optimization for map data is also proposed, which realizes realistic restoration of scene and good visualization effect. Due to no human involvement, automated processes can control human error and improve efficiency.

Power Equipment Segmentation Algorithm Based on Infrared and Visible Images Registration

LIU Xiao-kang, WAN Xi, TU Wen-chao, ZHOU Qing-kai, TIAN Zheng-wen

2020, 0(08): 26-30. doi:10.3969/j.issn.1006-2475.2020.08.005

Asbtract ( 348 )

PDF (5673KB) ( 263 )

References | Related Articles | Metrics

The fault of power equipment may cause large-scale power outages in substations, bringing out huge losses. According to the characteristics of heat generation during operation of power equipment, a power equipment segmentation method based on infrared and visible images registration is proposed to facilitate fault detection. Firstly, the structured random forest is used to detect the edge of infrared and visible images of the power equipment, and a multi-scale Gaussian pyramid of the visible edge image is constructed. Then the infrared and visible images are registered with the normalized mutual information. Finally, the Otsu threshold segmentation result of the infrared image is combined with the registration result to segment the power equipment in the visible image. The experimental results show that the proposed algorithm can register and segment accurately, and it has certain practicability.

A JPDA Multi-sensor Data Fusion Method for Association Probability Weighting

LIU Jian-feng,

2020, 0(08): 31-40. doi:10.3969/j.issn.1006-2475.2020.08.006

Asbtract ( 323 )

PDF (3040KB) ( 205 )

References | Related Articles | Metrics

Aiming at the problem that it is difficult to track multiple targets in complex environment of single sensor joint probabilistic data association(JPDA), this paper proposes a method of measurement-target association probabilistic statistical weighting parallel and sequential multi-sensor data fusion based on JPDA. Firstly, JPDA algorithm of single sensor is given. Then, the mathematical model of multi-sensor JPDA is given. Based on this model and using association probabilistic weighting, the parallel and sequential data fusion formulas are deduced, which have certain guiding significance for multi-sensor data fusion. Finally, the distance RMSE of target tracking is simulated for single sensor JPDA method under different clutter density, process and observation noise. The results show that the distance RMSE of target raise with the increasing of these three indicators. Simultaneously, the pedestrian tracking performance on data set PETS2009 is simulated for two kinds of multi-sensor JPDA methods of this paper and several other tracking methods. The results show that the parallel and the sequential multi-sensor JPDA methods of this paper are superior to other methods in tracking accuracy, tracking position accuracy, track maintenance and track loss. Furthermore, sequential fusion method is slightly better than parallel multi-sensor JPDA method in tracking performance.

Control Flow Error Detection Method Based on Multi-layer Segmented Labels

ZHENG Wei-ning, Zhuang Yi, GU Hao-wei

2020, 0(08): 41-50. doi:10.3969/j.issn.1006-2475.2020.08.007

Asbtract ( 236 )

PDF (1230KB) ( 223 )

References | Related Articles | Metrics

A control flow error detection method based on Multi-layer Segmented Labels (CFMSL) is proposed, which detects the control flow errors of the program online by updating and checking the multi-layer segmented labels. CFMSL can automatically embed label update and inspection instructions into the program during compile time, so as to realize dynamic inspection effect during program execution. The label design and calculation methods proposed in this paper are creative, which reduces the time and space complexity greatly, and has the abilities to handle complex programs and detect subtle control flow errors. CFMSL has the ability of batch and automation processing program through LLVM pass files. Finally, the fault injection tool designed in this paper is used to simulate the impact of control flow errors on software. At the same time, the error detection capabilities and overhead of CFMSL are evaluated. The experimental results show that compared with other methods, CFMSL has a lower overhead on the time and space and has higher error detection capabilities, which shows the superiority of the proposed method.

Method of Efficient Point Cloud Recognition Based on Attention Mechanism

LIN Qin-zhuang, HE Zhao-shui

2020, 0(08): 51-55. doi:10.3969/j.issn.1006-2475.2020.08.008

Asbtract ( 267 )

PDF (727KB) ( 181 )

References | Related Articles | Metrics

For point cloud recognition, methods of mapping point cloud data into two-dimensional pictures or restoring it to three-dimensional space have some shortcomings, such as large computation complexity and poor universality of scene. To address the problems, this paper proposes a method of deep residual learning network based on attention mechanism. The method obtains the weight distribution and key points of different points in the point cloud by the attention mechanism, and directly uses the point cloud data for efficient recognition. By the experiment, this paper studies and compares the recognition ability of different methods on the datasets MNIST and ModelNet40. The experimental results show that, compared with the methods respectively based on two-dimensional pictures, based on three-dimensional space and point cloud processing directly, the proposed method has the advantages of small parameter, small calculation and higher efficiency while ensuring high recognition accuracy.

An Improved Algorithm of Faster R-CNN Based on Variable Weight Loss Function and OHEM

SHI Fei, QIU Zhen, HAN Qin, LI Jin-geng, QIAN Hui-min, XIANG Wen-bo

2020, 0(08): 56-62. doi:10.3969/j.issn.1006-2475.2020.08.009

Asbtract ( 436 )

PDF (2019KB) ( 202 )

References | Related Articles | Metrics

Object detection algorithm based on deep convolutional neural network has become a research hotspot in the field of object detection, which includes two-stage object detection algorithm based on region proposal and one-stage object detection algorithm based on position regression. Faster R-CNN is one of the typical algorithms for two-stage object detection. However, the imbalance between simple examples and hard examples in the training data set and the inter-class imbalance of sample data are important reasons that affect the detection accuracy of Faster R-CNN. In this paper, an improved algorithm of Faster R-CNN based on variable weight loss function and OHEM is proposed. Specifically, the Focal Loss function is introduced into the classification part of the network to adjust the inter-class imbalance of sample data and improve the imbalance of the number of simple examples and the number of hard examples by adjusting the weight. At the same time, the network structure is modified, and online hard example mining is introduced to further balance the number of simple samples and the number of hard samples so as to improve the detection performance of the network. To verify the performance of the proposed algorithm, experiments on different data sets and different basic networks are conducted. The experimental results show that on the basic network VGG-16, the proposed algorithm improves the mAP by 09 percentage points on the Pascal VOC 2007 data set compared with the original algorithm and 1.7 percentage points on Pascal VOC 07+12 data set. On the basic network RES-101, the mAP of the proposed algorithm on Pascal VOC 2007 data set is 1.3 percentage points higher than that of the original algorithm, and the mAP of the proposed algorithm on Pascal VOC 07+12 data set is 1.5 percentage points higher.

Semi-supervised Overlapping Community Finding with Pairwise Constraints

XU Xiao-yuan, LI Hai-bo, YU Ben-cheng, LIU Fang

2020, 0(08): 63-68. doi:10.3969/j.issn.1006-2475.2020.08.010

Asbtract ( 271 )

PDF (959KB) ( 148 )

References | Related Articles | Metrics

In the research of finding overlapping community in complex networks, one of the ways to improve accuracy is by harnessing additional background information (e.g. from domain experts), which can be used as a source of constraints to guide the community detection process. In this paper, the potential of semi-supervised strategies to improve algorithms for finding overlapping communities in networks is explored. We introduce constraints that must link and cannot link into the initialization phase of the process and in the subcommunity extension process, and propose an approach named PC-GCE(Pairwise Constrained Greedy Clique Expansion) for finding overlapping communities by using a limited number of pairwise constraints and combing greedy strategies. A comparative experiment is conducted between the simulated network data and the current unconstrained locally extended overlapping community discovery algorithm (GCE), experimental results show that the PC-GCE can achieve better performance than GCE on finding overlapping communities, and as the increasing numbers of pairwise constraints, PC-GCE shows greater performance in the finding accuracy.

User-weighted Slope One Algorithm Based on Graph Embedding

ZHONG Zhi-song, PENG Qing-hua, WU Guang-chao

2020, 0(08): 69-75. doi:10.3969/j.issn.1006-2475.2020.08.011

Asbtract ( 280 )

PDF (1060KB) ( 155 )

References | Related Articles | Metrics

Aiming at the problem of low prediction accuracy of the traditional Slope One recommendation algorithm on sparse data set, this paper proposes a weighted Slope One algorithm based on graph embedding. This algorithm first establishes a correlation graph with time-aware user similarity as the edges’ weight, and obtains user eigen vectors based on the graph embedding of this graph. It then produces intra-class weighted Slope One recommendations using Canopy clustering. Additionally, to optimize the performance of the algorithm, we make an implementation based on the Spark computing framework. Experimental results demonstrate that, compared with the traditional weighted Slope One algorithm, the proposed algorithm has better recommendation effect and score prediction accuracy on both sparse data sets, explicit and implicit scoring data sets.

Study on Client-side Factors Affecting Large-scale Load Generation

JIN Wen-ming, YAN Shuo-yan, QIAN Ju,

2020, 0(08): 76-81. doi:10.3969/j.issn.1006-2475.2020.08.012

Asbtract ( 208 )

PDF (987KB) ( 92 )

References | Related Articles | Metrics

For large online software systems, it is often necessary to conduct effective large-scale load testing to ensure the reliability of their services. However, the existing research mainly focuses on the server-side performance, client-side load generation models and client-side resource allocation, little work concerns on the client-side load generation ability. There lack detailed analysis and studies on the factors that may affect client-side load generation to guide large-scale load generation. To this end, this paper experimentally studies the factors affecting client-side load generation from two perspective: load generation mechanism and test cluster construction, comes to a set of data conclusions for large-scale load generation based on capability indicators. This paper provides data reference for testers to generate large-scale load on the client, which may help reduce the cost of large-scale load testing.

Networking Method for SDMANET Based on Multi-mode Radio

DAI Song, SUN Yan-tao, LIU Qiang, JIA Ze-qun,

2020, 0(08): 82-88. doi:10.3969/j.issn.1006-2475.2020.08.013

Asbtract ( 311 )

PDF (1156KB) ( 161 )

References | Related Articles | Metrics

Software-defined networking (SDN) is developing rapidly in a variety of network scenarios, including wired networks and data center networks. However, the use of SDN in Mobile Ad Hoc Networks (MANET) is still in its infancy. Due to the frequent topology changes, limited resources, and the use of distributed networking in MANET networks, applying SDN in MANET becomes more challenging. To this end, this paper proposes a Software-defined Mobile Ad Hoc Networks （SDMANET） networking method based on multi-mode radios. This method uses the dominating set algorithm to calculate the backbone nodes. The backbone nodes use out-of-band channels to communicate with the SDN controller, and perform dynamical allocation of TDMA time slots based on the backbone nodes at the MAC layer. Experimental results show that compared with OLSR protocol and direct out-of-band control method, this method has lower network control overhead and channel access delay, and has better performance in large-scale MANET networks.

Forecasting Method of Runoff in Dry Season of Rivers Based on SSA-PPR Model

HU Xin

2020, 0(08): 89-93. doi:10.3969/j.issn.1006-2475.2020.08.014

Asbtract ( 260 )

PDF (742KB) ( 133 )

References | Related Articles | Metrics

The real-time variability of river dry season runoff affects the accuracy of the prediction results, in order to obtain accurate prediction results and improve prediction efficiency, a method of predicting river dry season runoff change based on SSA-PPR model is proposed. This paper uses SSA-PPR model to build a big data statistical analysis model for river runoff change prediction in dry season, uses quantitative statistical feature analysis method to mine the dynamic change characteristics of runoff, and obtains the change statistical feature quantities, and combines the fuzzy information mining and adaptive learning to get the dynamic analysis results of river runoff change in dry season. According to the analysis results, the dynamic classification and recognition of the flow change are carried out, and the accurate prediction of the river runoff change in dry season is completed. The simulation results show that the prediction result of this method has higher accuracy, better adaptability and higher prediction efficiency, which effectively improves the convergence of the prediction process, and has a good guiding significance for the quantitative analysis of river runoff changes in dry season.

Decision-making Method for Airport Taxi Drivers Based on Deep Reinforcement Learning

WANG Peng-yong, CHEN Gong-tao, ZHAO Jiang-shuo

2020, 0(08): 94-99. doi:10.3969/j.issn.1006-2475.2020.08.015

Asbtract ( 354 )

PDF (1148KB) ( 250 )

References | Related Articles | Metrics

In order to deal with the difficulty of taxi dispatching in large transportation hub, especially in airport, from the view of the taxi driver’s profit, this paper proposes a decision-making method based on improved deep reinforcement learning. Firstly, the airport environment and the urban environment where the airport is located are simulated, and the driver’s states, actions, the rewards obtained from interaction with the environment and the state transitions are defined. Then, the states of the driver, as inputs, are fed into DQN to fit the values of Q-value function. Finally, through continuously simulating the drivers’ decisions by ε-greedy strategy and reward functions, this paper reaches the purpose of upgrading the parameters of DQN. The experiment results show that drivers can quantitatively get expected benefit for current decision actions and make proper decision through the model in simulated large, medium and small cities and other environments, so as to automatically complete the process of taxi dispatching.

Meteorological Data Storage and Retrieval System Based on MongoDB

CHEN Hao, ZHANG Ya, LUO Xi-chang, ZHANG Ya-li, LIU Wen-jing

2020, 0(08): 100-104. doi:10.3969/j.issn.1006-2475.2020.08.016

Asbtract ( 408 )

PDF (1209KB) ( 229 )

References | Related Articles | Metrics

In recent years, meteorological data shows a trend of multi-source and explosive growth, the traditional relational database can not meet the needs of meteorological data development. In combination with the geospatial characteristics of meteorological data, a MongoDB-based meteorological data storage and retrieval system is proposed. The system establishes spatial index of meteorological data, which speeds up the query efficiency of meteorological data and provides a strong support for fine and lattice forecast. The experimental results show that MongoDB has strong storage and retrieval capability for massive meteorological data, and its performance in all aspects is obviously better than that of relational database.

Design and Application of Big Data Platform of Anhui Meteorological Service for Agriculture

XU Jian-peng, ZHANG Hui, WU Qiong, WANG Hui, WANG Bing

2020, 0(08): 105-108. doi:10.3969/j.issn.1006-2475.2020.08.017

Asbtract ( 282 )

PDF (2289KB) ( 188 )

References | Related Articles | Metrics

Based on the needs of Anhui Meteorology for agricultural information services, such as personalized, precise, and intelligent, a big data platform of Anhui meteorological service for agriculture is developed using Hadoop architecture, natural language processing, correlation analysis, big data visualization and other related big data and artificial intelligence technologies. The big data platform for agricultural services brings together agricultural data resources from multiple departments in Anhui Province. Through the establishment of user behavior portraits and the prediction of network service hotspots, it provides accurate service product relevance recommendation services, and tracks and evaluates the network service effect of information service products, guides the development and production of follow-up key agricultural recommendations and decision-making service products, and at the same time establishes a big data display system of meteorological services for agriculture, explores the transformation of meteorological information services for agriculture from “people looking for information” to “information looking for people”. The platform has been applied in Anhui Meteorological services for agricultural business, which has improved service capabilities and has good industry reputation and social influence.

Dynamic Suppression Algorithm of Semi-distributed Zombie Network Based on Big Data

LIU Zhang-rong

2020, 0(08): 109-113. doi:10.3969/j.issn.1006-2475.2020.08.018

Asbtract ( 212 )

PDF (958KB) ( 246 )

References | Related Articles | Metrics

In order to improve the security of semi-distributed botnet, a dynamic suppression algorithm of semi-distributed botnet based on big data is proposed. The method of baud interval equalization control is used to compensate the dynamic characteristics of the semi-distributed botnet. The sampling model of the dynamic characteristics of the semi-distributed botnet is constructed. The decision equalization method is used to analyze the collected dynamic characteristics quantitatively and recursively and extract the statistical characteristics of the semi-distributed botnet. On this basis, the big data optimization calculation method is used to achieve the true optimal suppression of the semi-distributed botnet. Under the embedded environment, the optimal inhibition parameters of the semi-distributed botnet are combined with the results of load response to realize the dynamic inhibition of the semi-distributed botnet. The simulation results show that the semi-distributed botnet dynamic suppression method has a good effect, it improves the suppression precision, shortens the suppression time, and reduces the bit error rate of network output.

Service Vulnerability Mining of Android System Based on Genetic Algorithm

ZHANG Zhi-wei, GAN Gang

2020, 0(08): 114-121. doi:10.3969/j.issn.1006-2475.2020.08.019

Asbtract ( 339 )

PDF (1149KB) ( 214 )

References | Related Articles | Metrics

In order to solve the problem of low efficiency in mining service vulnerabilities in Android system by conventional fuzzy testing, this paper proposes and implements a framework for mining service vulnerabilities in Android system based on genetic algorithm, named ASFuzzer. The framework uses Binder driver to interact with system services to send test cases to the target. According to the feedback of the test results, the genetic algorithm is guided to continuously change the test parameters, and an efficient genetic selection operator model based on probability sorting and combination is proposed to improve the sample coverage and fuzzy test efficiency. Through the testing of the framework on mobile phones of different system versions, multiple system service vulnerabilities are discovered. Compared with the traditional fuzzy testing method, the experimental results show that the scheme has more advantages in the efficiency of vulnerability mining.

Model Construction of College Network Security System Based on SSL Virtual Technology

DAI Rui-feng

2020, 0(08): 122-126. doi:10.3969/j.issn.1006-2475.2020.08.020

Asbtract ( 286 )

PDF (1547KB) ( 285 )

References | Related Articles | Metrics

Researching on the relevant theories of SSL protocol, based on the construction of campus security information network system, a university network security system model based on SSL virtual technology of PKI identity authentication system is designed. In order to verify the security and effectiveness of the model system, a system simulation environment is constructed. Through simulation analysis, it shows that the system has good security and practicability, and various functional modules can be smoothly implemented, which can ensure the confidentiality of communication data in client side and server side. The model proposed in this paper has a certain reference significance for the construction of university network information system.

Table of Content