Table of Content

31 January 2022, Volume 40 Issue 1

Previous Issue    Next Issue

Special Issue on Computer Applications

Emotional Analysis of Brain Waves Based on CNN and Bi-LSTM

ZHU Li, YANG Qing, WU Tao, LI Chen, LI Ming

2022, 40(1):  1-12.  doi:10.3969/j.issn.0255-8297.2022.01.001

Asbtract ( 2357 )   PDF (1792KB) ( 402 )  

References | Related Articles | Metrics

Aiming at the problem that most emotion recognition methods rely on manual feature extraction, a hybrid model based on convolutional neural network (CNN) and bidirectional long short-term memory (Bi-LSTM) network is proposed. Firstly, one-dimensional data is converted into two-dimensional data, and spatial features are extracted by CNN. Then the one-dimensional data is input into Bi-LSTM to obtain temporal features. Finally, the fused spatial and temporal features are input into Softmax classifier to obtain final classification results. Experimental results on DEAP dataset show that CNN and BiLSTM hybrid model has good classification performance, and the accuracy in potency and arousal reaches 88.55% and 89.07%, respectively, proving the proposed model is a feasible and affective EEG emotion classification model.

Behavior Imitation Robotic System with Cognition Capacity

BAO Zhenshan, DING Yilong, ZHANG Wenbo

2022, 40(1):  13-24.  doi:10.3969/j.issn.0255-8297.2022.01.002

Asbtract ( 1825 )   PDF (4503KB) ( 135 )  

References | Related Articles | Metrics

Behavioral imitation is one of the important technologies for robots to show their intelligence. How to make the behaviors and actions imitated by robots similar to the demonstrating actions of human has become a hot research topic. In this paper, we design an improved robot behavior modeling framework based on simple method. The framework collects teaching action using normal monocular camera, and introduces behavior semantic recognition module and key action extraction module into the simple method. The framework enables robots to understand instructor's behavior semantics and then imitate instructor's behaviors. Finally, this framework is deployed on the HBE-ROBONOVA-AI II humanoid robot platform, and experiments are conducted using independently collected single-person action video data as input. Compared with the experimental results of other mainstream frameworks, this framework works with more excellent comprehensive performance in three aspects of accuracy, balance and similarity, and demonstrates a unique cognitive ability to instructor's behaviors.

Static Multimodal Sentiment Analysis of Online Reviews

WANG Kaixin, XU Xiujuan, LIU Yu, ZHAO Zhehuan, ZHAO Xiaowei

2022, 40(1):  25-35.  doi:10.3969/j.issn.0255-8297.2022.01.003

Asbtract ( 1875 )   PDF (1977KB) ( 165 )  

References | Related Articles | Metrics

This paper proposes a static multi-modal sentiment classification model based on Pre-LN Transformer. This model firstly extracts semantic features from reviews using the encoder in Pre-LN Transformer structure, in which the multi-head self-attention mechanism allows the model to learn relevant emotional information in different subspaces. Then our model extracts the image features according to ResNet in the reviews. On the basis of feature level fusion, the visual attention mechanism guides the sentiment classification of text and realizes the static multimodal sentiment analysis of online reviews. Experimental results show that our model improves the performance by 1.34% and 1.10% in evaluation accuracy than BiGRU-mVGG and Trans-mVGG on Yelp datasets, which verifies the effectiveness and feasibility of the proposed model.

Fine-Grained Image Classification Based on Inference Graph of Attention Network

ZHENG Zhiwen, GAN Jianhou, ZHOU Juxiang, OUYANG Zhaoxiang, LU Zeguang

2022, 40(1):  36-46.  doi:10.3969/j.issn.0255-8297.2022.01.004

Asbtract ( 1840 )   PDF (4758KB) ( 152 )  

References | Related Articles | Metrics

Aiming at the task of fine-grained classification of scene images, this paper proposes a fine-grained image classification method based on the attention network inference graph by integrating the multimodal information of image visual and textual features. First, we extract the global visual feature, local visual features and text features of the scene image, and form a new splicing feature by embedding the position information into the local visual features and textual features respectively. The feature is accordingly used as a node of the graph structure to generate a heterogeneous graph. Then, we design two meta-paths to decompose the heterogeneous graph into two isomorphic graphs, and put them into a two-level attention network inference graph with node-level attention and semantic-level attention. Finally, richer fine-grained feature expression can be obtained by multimodal fusion operations with the output node features and global visual feature. The proposed model enables effective combination of multimodal fusion and graph attention network, and performs strong competitiveness comparing with the current advanced mainstream methods on the two scene text fine-grained image datasets of Con-Text and Drink Bottle.

Modeling of Multi-agent City Safety and Livability Based on Street Perspective

PAN Lihu, YANG Fenyu, LU Feiping, QIN Shipeng

2022, 40(1):  47-60.  doi:10.3969/j.issn.0255-8297.2022.01.005

Asbtract ( 1803 )   PDF (4740KB) ( 102 )  

References | Related Articles | Metrics

In order to solve the problem of urban safety livability under the influence of environmental and human activities, this paper builds a multi-agent model of city safety livability based on the data of ten streets in Futian District, Shenzhen. The model is realized by integrating geographic information system (GIS) with Repast simulation platform. With the model, we have simulated the evolution of urban safety development in Futian District in the next 20 years from the perspective of streets, and analyzed the dynamic interactive feedback mechanism of safety livability, resident satisfaction, and family relocation behavior of each street in Futian District. The simulation experiment proves that the multi-agent model is effective for predicting urban development under the influence of multiple factors.

Deep-Level Kernel Hook Mining Algorithm and Its Application in Software Security

LU Dengkai, YU Yongbin, YU Wenjian, TANG Qian, LIANG Shouyi

2022, 40(1):  61-68.  doi:10.3969/j.issn.0255-8297.2022.01.006

Asbtract ( 1720 )   PDF (967KB) ( 71 )  

References | Related Articles | Metrics

This paper studies the protection principle of kernel hooks in the Windows operating system and proposes a deep-level kernel hook mining algorithm to solve the shortcomings of the interactive disassembler professional (IDA) cross-reference function. Firstly, the algorithm is used to dig out the internal calls of specified kernel functions and all the called positions of the kernel functions containing hooks. Then, we use Python to write mining algorithms based on the principle of function calls. Finally, we use C++ to write a driver program for passing-protection experiment. The performance of overprotection experiment is successful, which proves the effectiveness of the mining algorithm and the comprehensiveness of mining results.

Ensemble Classification Algorithm Based on Cost Sensitive Convolutional Neural Networks

ZHOU Chuanhua, XU Wenqian, ZHU Junjie

2022, 40(1):  69-79.  doi:10.3969/j.issn.0255-8297.2022.01.007

Asbtract ( 2082 )   PDF (623KB) ( 246 )  

References | Related Articles | Metrics

Aiming at the problem of low recognition rate of a few types of samples in unbalanced data sets, a classification algorithm based on cost sensitive convolutional neural network and AdaBoost (AdaBoost-CSCNN) was proposed. The cost sensitive convolutional neural network (CSCNN) is constructed by coordinating the cross entropy loss function of convolutional neural network (CNN) with a specific cost sensitive index. In training process, cost weighting mechanism is used to reduce the loss degree of key feature attributes of a few samples and realize the classification effect of a single CSCNN as a base classifier in AdaBoost. To verify the effectiveness of the algorithm, we carried out experiments on 9 data sets with different imbalance rates. Experimental performances, including Accuracy, Recall, F1-score and AUC, show that the AdaBoost-CSCNN algorithm has a good display for unbalanced data set classification.

Infrared Image Fusion Based on NSCT and Compressed Sensing

JIN An'an, LI Xiang, ZHANG Li, XIONG Qingzhi

2022, 40(1):  80-92.  doi:10.3969/j.issn.0255-8297.2022.01.008

Asbtract ( 1801 )   PDF (11751KB) ( 91 )  

References | Related Articles | Metrics

Aiming at the problems of low quality, lack of information and non-prominent edge details in the fusion process of infrared and visible images, this paper proposes a compressed sensing image fusion and reconstruction algorithm based on non-subsampled contourlet transform (NSCT) and sparse representation. Firstly, a source image is decomposed by using NSCT to obtain corresponding high-frequency sub-band and low-frequency sub-band images. Then, the high-frequency sub-band images are fused by using the highfrequency fusion rules based on compressed sensing to obtain high-frequency fusion coefficients. For the low-frequency sub-band images, low-frequency fusion coefficients are obtained by using the low-frequency fusion rules based on dictionary learning. Finally, a fusion image is obtained by using the inverse NSCT transformation to achieve superresolution recovery of infrared and visible images. Experimental results show that the images fused by this algorithm have good performance in metrics, such as average gradient, edge intensity, information entropy, edge information retention and spatial frequency, and prove that this fusion algorithm has significant advantages in image fusion quality.

Mask Wearing Detection in Complex Scenes Based on Mask-YOLO

WEI Mingjun, ZHOU Taiyu, JI Zhanlin, ZHANG Xinnan

2022, 40(1):  93-104.  doi:10.3969/j.issn.0255-8297.2022.01.009

Asbtract ( 2194 )   PDF (22008KB) ( 451 )  

References | Related Articles | Metrics

Aiming at the problem of low detection accuracy caused by occlusion, density and small scale in mask wearing detection in public places, a Mask-YOLO algorithm is proposed based on real-time target detection algorithm YOLOv3. First, the algorithm introduces channel attention mechanism in the process of feature fusion, effectively highlights the important features, reduces the influence of redundant features after fusion, and effectively improves the feature utilization. Then, complete intersection over union (CIoU) loss is used instead of mean square error (MSE) as the loss function of frame regression to improve the positioning accuracy. Finally, in addition to the cases of detecting wearing and not wearing masks, incorrect wearing of masks is also detected. Experimental results show that Mask-YOLO algorithm improves mean average precision (mAP) by 4.78% when frame per second (FPS) decreases by only 1% compared with YOLOv3 algorithm. As compared with other mainstream target detection algorithms, Mask-YOLO algorithm also has better detection effect and robustness for mask wearing detection in complex scenes.

Segmentation Model of COVID-19 Lesions Based on Triple Attention Mechanism

LEI Qianhui, PAN Lili, SHAO Weizhi, HU Haipeng, HUANG Yao

2022, 40(1):  105-115.  doi:10.3969/j.issn.0255-8297.2022.01.010

Asbtract ( 1934 )   PDF (3285KB) ( 159 )  

References | Related Articles | Metrics

In order to solve the problem of low intensity contrast between infected areas and normal tissues, A corona virus disease 2019 (COVID-19) segmentation model TMNet is proposed based on triple attention mechanism (TAM), and applied to conditional generative adversarial network in this paper. The MultiConv module in TM-Net can automatically extract rich features of infected areas in lung slices. These features contain different types of lesion information. The designed TAM, which integrates spatial, channel and positional attention modules, can accurately locate lesions in the infected area. By composing of three types of loss functions, the loss function of TM-Net can minimize the differences between prediction graphs and real labels, thus optimizing the TM-Net. Experiment and evaluations conducted on COVID-19 data sets show that the average dice similarity coefficient (DSC) of ground glass opacities (GGO) and consolidation of TM-Net are 1.4% and 0.5% higher than the results of attention U-Net and R2U-Net, respectively, proving the accuracy improvement of TM-Net in COVID-19 lesions segmentation.

Sparrow Search Algorithm Based on Levy Flight Disturbance Strategy

MA Wei, ZHU Xian

2022, 40(1):  116-130.  doi:10.3969/j.issn.0255-8297.2022.01.011

Asbtract ( 2147 )   PDF (817KB) ( 192 )  

References | Related Articles | Metrics

In order to solve the problems of insufficient search diversity in late iteration and easy falling of local optimization in traditional sparrow search algorithm, an improved sparrow search algorithm (ISSA) based on Levy flight disturbance strategy is proposed. Firstly, the algorithm uses Sin chaos search mechanism to improve population initialization strategy. Secondly, in the process of sparrow population foraging search, Levy flight disturbance mechanism is introduced to drag the appropriate step of population movement, and the diversity of spatial search is then increased. Finally, experiment on 14 typical highdimensional test functions has been carried out, and the results show that compared with the traditional sparrow search algorithm and two other recently proposed chaos sparrow search algorithm (CSSA) and ISSA, the proposed algorithm in this paper can effectively avoid the search process falling into local optimization, and achieve high optimization rate and strong convergence ability, and shows feasibility in solving problems of multi-peak and high-dimensional space optimization.

Remote Sensing Image Object Detection Based on MFANet and Contextual Features Fusion

WANG Peng, ZHENG Wenfeng, SHI Jin, JIN Shuo, LIU Zihao

2022, 40(1):  131-144.  doi:10.3969/j.issn.0255-8297.2022.01.012

Asbtract ( 1774 )   PDF (12424KB) ( 158 )  

References | Related Articles | Metrics

Remote sensing images have the characteristics of complex background, large variations of object sizes and inter-class similarity, which lead to poor object detection results. An effective and robust remote sensing image object detection method based on Faster R-CNN is proposed. First, we introduce deformable convolution, feature modulation mechanisms and dilated convolution to construct a modulated feature adaptation network named MFANet, which can extract more accurate and complete object information. Second, a contextual feature pyramid network named CFPN is introduced to exploit richer and more discriminative feature representations. CFPN can solve the problems of insufficient high-level semantic information in the process of feature transfer and lack of effective communication between multi-size receptive fields. Finally, complete IoU (CIoU) loss is introduced into bounding box regression to further improve the accuracy of object detection. To verify the validity of the proposed method, we conduct experiments on public datasets DIOR, RSOD, and NWPU VHR-10. Experimental results show that compared with the Faster R-CNN with FPN method, IF-RCNN obtains an absolute gain of 8.43%, 7.5% and 8.0% in the average detection accuracy on the three datasets, respectively, which suggests that our proposed method is more effective and robust.

KNN-GWD Recommendation Model and Its Application

JI Deqiang, WANG Hairong, CHE Miao, WANG Jiaxin

2022, 40(1):  145-154.  doi:10.3969/j.issn.0255-8297.2022.01.013

Asbtract ( 1667 )   PDF (652KB) ( 100 )  

References | Related Articles | Metrics

In order to solve the problem of poor accuracy in traditional recommendation, a multi-layer K-nearest neighbor (KNN) network recommend model KNN-GWD, in combination of graph neural network (GNN) and wide & deep network was constructed. In the model, the KNN classification method is for data noise filtering to improve data quality. GNN is used to extract the node embedding representation of user's conversation graphs, and capture user's short-term interest by weighting user's global characteristics based on attention mechanism. Wide&Deep is used to solve the problem of model overgeneralization in the case of sparse data. In order to verify the effectiveness of the model, comparative experiments were carried out on MovieLens-1M, Bing-News and Book-Crossing data sets with this model and six other traditional recommendation methods. Experimental results show that the evaluation indicators of this model are better. In order to further verify the feasibility of the proposed model in the actual application field, an agricultural integrated management App fertilizer recommendation system was built with the accuracy of recommended results of 0.721 and the area under curve of 0.784, which met the expected application requirements.

Track Slab Crack Detection Method Based on TSCD Model

LI Wenju, ZHANG Yaoxing, CHEN Huiling, LI Peigang, SHA Liye

2022, 40(1):  155-166.  doi:10.3969/j.issn.0255-8297.2022.01.014

Asbtract ( 1797 )   PDF (2650KB) ( 91 )  

References | Related Articles | Metrics

In order to solve the problem of track slab crack detection, a track slab crack detection model based on branch cascaded convolutional neural network, TSCD, is proposed. First, the model highlights the position information of track slab cracks through attention mechanism and structure of search branches to suppress interference information. Second, it realizes the pixel-level detection of cracks by structure of detecting branches. Finally, in order to solve the problem of image detail degradation in detection results, parameter mapping is used to achieve up-sampling of the feature maps. Experimental results show that the proposed model in this paper can not only detect the cracks in track plate surface images accurately with pixel accuracy rate of 97.56% and F1-score of 86.28%, but also performs strong generalization in cross-dataset tests.

Research on UAV Detection and Counter Technologies for Security in Key Areas

JIANG Dongting, FAN Changjun, YONG Qirun, QU Chongxiao, LIU Shuo, ZHANG Yongjin

2022, 40(1):  167-178.  doi:10.3969/j.issn.0255-8297.2022.01.015

Asbtract ( 1812 )   PDF (630KB) ( 219 )  

References | Related Articles | Metrics

This study analyzes the characteristics of unmanned aircraft vehicle (UAV), security risks of UAV and difficulties in counter unmanned aircraft vehicle (C-UAV). UAV detection techniques including radar, radio electro-optical, and acoustic sensors as well as UAV interdiction techniques including RF/GNSS jamming, spoofing, laser, nets and so on have been thoroughly studied in this paper. The market application of these technologies is analyzed, and the advantages and disadvantages of these technologies are compared and analyzed. Finally, some suggestions on UAV defense and control systems in various key areas are provided.