Computer and Modernization

Aspect Based Sentiment Analysis Model Based on Knowledge Enhancement

LI Shi-yue, MENG Jia-na, YU Yu-hai, LI Xue-ying, XU Ying-ao

2023, 0(10): 1-8. doi:10.3969/j.issn.1006-2475.2023.10.001

Asbtract ( 225 )

PDF (2224KB) ( 78 )

References | Related Articles | Metrics

Aspect based sentiment analysis can accurately determine the emotional polarity of aspect words in sentences， and plays an important role in social networking， e-commerce and other fields. Most of the existing methods model the relationship between context and target words through sequence representation or attention mechanism， but ignore the background knowledge of text and the conceptual links between aspect words， resulting in insufficient semantic relationships learned. To solve the above problems， the Aspect Based Sentiment Analysis Model Based on Knowledge Enhancement （ABSA-KE） is proposed. First， the features are extracted and the corresponding word vector is obtained through the pre-training model BERT， and the dependency tree corresponding to the text is obtained using the parser. Then， the joint modeling of BiLSTM and graph attention network is used to learn the node embedded representation and obtain the text vector. Second， the external knowledge base is used to introduce the aspect word knowledge vector in different contexts to enhance the aspect level emotion analysis model， and finally the emotion classification task is carried out. Compared with the existing models， the experimental results show that the proposed model is effective and reasonable in aspect level emotion analysis tasks.

Sequence Recommendation Model Based on Dynamic Convolution and Self-attention

ZHENG Hai-li, CHEN Ping-hua

2023, 0(10): 9-16. doi:10.3969/j.issn.1006-2475.2023.10.002

Asbtract ( 127 )

PDF (3775KB) ( 51 )

References | Related Articles | Metrics

Sequence recommendation dynamically models user interests according to the historical interaction records of users and items， and recommends next item. The sequence modeling user interests is usually divided into long-term interest dependency and short-term interest dependency. The existing methods either divide the sequence according to the interaction order， respectively model the long-term and short-term interest dependence， separately model the user interest， or extract the features of the same interactive sequence in parallel with different feature extraction technologies to obtain the global and local interest representation， ignoring the fact that the user intention at different times exists in the behavior context at that time. This paper proposes DConvSA to model dynamic interest by using dynamic convolution and self-attention. Dynamic convolution is used to extract local dynamic interest， and convolution kernel is generated according to different context items to adaptively calculate the importance of items. Combined with the self-attention mechanism， the overall significant item dependency is obtained. At the same time， the global and local interest dependencies at each time are fused in an explicit way to better model the relationship between interests at different times. Experiments are conducted on three public datasets， using recall rate， average reciprocal ranking and normalized cumulative gain for performance evaluation. The results show that the recall rate， mean reciprocal ranking and normalized discounted cumulative gain increased by at least of 1.53%， 3.77% and 3.28% on the MovieLens-1M dataset， 1.86%， 1.94% and 2.46% on the Amazon Beauty dataset， and 0.22%， 0.97% and 1.08% on the Steam dataset.

Feature-level Multimodal Fusion for Depression Recognition

GU Ming-xuan, FAN Bing-bing

2023, 0(10): 17-22. doi:10.3969/j.issn.1006-2475.2023.10.003

Asbtract ( 244 )

PDF (1213KB) ( 99 )

References | Related Articles | Metrics

Abstract: Depression is a common psychiatric disorder. However， the existing diagnostic methods for depression mainly rely on scales and interviews with psychiatrists， which are highly subjective. In recent years， researchers have devoted themselves to identifying depressed patients by EEG features or audio features， but no study has effectively combined EEG information with audio information， ignoring the correlation between audio and EEG data. Therefore， this study proposes a feature-level multimodal fusion model to improve the accuracy of depression recognition. We combine the audio and EEG modality information based on a fully connected neural network. Our experiments show that the accuracy of depression recognition using feature-level multimodal fusion model on the MODMA dataset reaches 81.58%， which is higher than that of using single-modality. The results indicate that the feature-level multimodal fusion model can improve the accuracy of depression recognition compared to single-modality. Our research provides a new perspective and method for depression recognition.

An Improved Sparrow Search Algorithm Based on Multi-strategy

LU Lei, HE Zhi-ming, HUANG Zhi-cheng

2023, 0(10): 23-31. doi:10.3969/j.issn.1006-2475.2023.10.004

Asbtract ( 111 )

PDF (2714KB) ( 58 )

References | Related Articles | Metrics

To address the problems that the population diversity of the sparrow search algorithm （SSA） decreases in the late iteration and easily falls into local optimum， a multi-strategy based improved sparrow search algorithm （MUSSA） is proposed. Firstly， MUSSA uses opposition-based learning and iterative strategy to enhance population diversity. According to the forgotten decline strategy， the number of populations using the reverse iteration strategy is gradually reduced， the loss of useless search is reduced， and the convergence speed of the algorithm is accelerated. Then， the adaptive weight spiral search strategy and reference frame mechanism are introduced to modify the discoverer update formula， further expand the search range of individuals and enhance the global search capability of the algorithm. Finally， direction factor and non-static selection strategy are introduced into follower renewal strategy to enhance local mining excavation. The simulation results of 13 benchmark test functions show that MUSSA has better optimization performance than SSA， HHO， WOA and AO.

Breast Cancer Prediction and Feature Analysis Model Based on CatBoost and SHAP

JIA Xiao-yao,

2023, 0(10): 32-38. doi:10.3969/j.issn.1006-2475.2023.10.005

Asbtract ( 106 )

PDF (3562KB) ( 58 )

References | Related Articles | Metrics

To address the problems of insufficient performance and poor interpretability of current breast cancer prediction models， this paper proposes a breast cancer prediction and feature analysis model incorporating CatBoost and SHAP. First， the original breast cancer dataset is processed with outliers and data normalization to improve the quality of the data. Then， a model for breast cancer prediction based on CatBoost is built and generalization ability analysis is performed. Finally， the prediction model is combined with SHAP for interpretable analysis to explore the key factors affecting breast cancer. The model is validated using the Breast Cancer Wisconsin （Diagnostic） dataset from the University of Wisconsin， and the results show that the Accuracy value of 99.30%， Precision value of 99.50%， Recall value of 98.91%， and F1 value of 99.19% are better than the existing literature. The superiority of this model is verified by the fact that the Accuracy index improved by 1.12-6.90 percentage points， the Precision index improved by 2.00-7.50 percentage points， the Recall index improved by 2.41-6.91 percentage points， and the F1 value improved by 2.19-7.19 percentage points. In addition， the SHAP model yields the core factors affecting breast cancer， such as concave points_worst， perimeter_worst， and area_worst， which provide the principle support for doctors’ diagnosis.

Health Event Prediction Model Based on Dynamic and Static Features of Graph Nodes

CHEN Jun-yi

2023, 0(10): 39-44. doi:10.3969/j.issn.1006-2475.2023.10.006

Asbtract ( 59 )

PDF (1602KB) ( 49 )

References | Related Articles | Metrics

With the wide application of electronic healthcare records （EHRs）， the prediction of clinical health events based on deep learning has attracted the attention of many researchers. The existing work mainly focuses on mining the higher-order temporal characteristics of patients， and fails to effectively learn the hidden relationship between diseases. Aiming at the problem of disease representation learning， this paper proposes a novel disease representation model （Health Event Prediction Model Based on Dynamic and Static Features of Graph Nodes， DuDas）， through which the final hidden representation of disease mined by the model contains static and dynamic information， and finally realizes the prediction of clinical tasks. Firstly， the disease graph is constructed according to the disease co-occurrence frequency， and the initial hidden representation is assigned to each disease node by the one-hot coding module. Then， the static representation of the disease is excavated according to the static mining module， and it is fused with the corresponding initial hidden representation as the initial dynamic hidden representation. The dynamic relationship between diseases is mined according to the graph convolution module to learn the final dynamic hidden representation of disease nodes. Due to the temporal nature of patient visits， this article uses gated circulation units to mine the relationship between historical diagnostic information and current diagnostic information. In order to verify the effectiveness of the proposed method， we perform experimental verification on two real datasets. Experimental results show that the proposed model in this paper reaches the higher level in the task of predicting health events.

Image Super-resolution Reconstruction Based on Spatial Attention Residual Network

XING Shi-shuai, LIU Dan-feng, WANG Li-guo, PAN Yue-tao, MENG Ling-hong, YUE Xiao-han

2023, 0(10): 45-52. doi:10.3969/j.issn.1006-2475.2023.10.007

Asbtract ( 115 )

PDF (3939KB) ( 66 )

References | Related Articles | Metrics

Hierarchical features extracted from convolutional neural networks contain affluent semantic information and they are crucial for image reconstruction. However， some existing image super-resolution reconstruction methods are incapable of excavating detailed enough hierarchical features in convolutional network. Therefore， we propose a model termed spatial attention residual network （SARN） to relieve this issue. Specifically， we design a spatial attention residual block （SARB）， the enhanced spatial attention （ESA） is embedded into SARB to obtain more effective high-frequency information. Secondly， feature fusion mechanism is introduced to fuse the features derived from each layer. Thereby， the network can extract more detailed hierarchical features. Finally， these fused features are fed into the reconstruction network to produce the final reconstruction image. Experimental results demonstrate that our proposed model outperforms the other algorithms in terms of quantitative evaluation and visual comparisons. That indicates our model can effectively utilize the hierarchical features contained in the image， thus achieve a better super-resolution reconstruction performance.

Apple Defect Detection Algorithm Based on NAM-YOLO Network

ZHANG Jia-Qi, XU Qi-lei

2023, 0(10): 53-58. doi:10.3969/j.issn.1006-2475.2023.10.008

Asbtract ( 113 )

PDF (4281KB) ( 69 )

References | Related Articles | Metrics

Aiming at the problems of apple defect detection， such as frequent false detection， leakage detection and easy confusion of defects， we propose an apple defect detection algorithm based on improved YOLOv5. Apple defect detection is very important for apple sorting. The existing methods of apple defect detection mainly extract color and texture features through machine learning or convolutional neural network， but there are problems such as error detection， missing detection and insufficient feature extraction ability. It can not meet the requirements of accuracy and real-time defect detection. NAM-YOLO algorithm mainly has three core ideas: 1） By adding TRANS module to the backbone network， features and global information can be better integrated； 2） The weighted bidirectional feature pyramid network is used to fuse features of different scales； 3） The NAM attention mechanism based on normalization is introduced into the neck network to strengthen the key features of the target region and improve the detection accuracy of the network. Experimental results show that the mAP of the improved algorithm reaches 98.90% and the accuracy is 98.73%. Compared with other models， this model has better feature fusion ability and can better meet the actual needs of apple sorting.

Safety Helmet Detection Based on Lightweight YOLOv5

LI Yan-man, WANG Bi-heng, ZHAO Ling-yan

2023, 0(10): 59-64. doi:10.3969/j.issn.1006-2475.2023.10.009

Asbtract ( 145 )

PDF (5102KB) ( 68 )

References | Related Articles | Metrics

There is a large amount of data in the intelligent monitoring system of distribution network， which objectively requires the algorithm to have high real-time performance. Based on this， the YOLOv5 algorithm is improved in light weight， including improving the K-means algorithm clustering anchor box， using the Hard-swish activation function and the CRD loss function， and at the same time integrating the ShuffleNet structure in the backbone network and adopting the Attention mechanism in the FPN module. The presented model， SNAM-YOLOv5 （ShuffleNet and Attention Mechanism-You Only Look Once version 5）， can significantly improve the detection performance and the processing speed of small targets and occluded targets. The results of safety helmet detection based on HiSilicon Hi3559A embedded platform show that the model is superior to similar algorithms and has good real-time performance.

HRNet Image Semantic Segmentation Algorithm Combined with Attention Mechanism

YE Si-jia, WEI Yan, DU Han-yu, DENG Jin-zhi

2023, 0(10): 65-69. doi:10.3969/j.issn.1006-2475.2023.10.010

Asbtract ( 89 )

PDF (2303KB) ( 50 )

References | Related Articles | Metrics

Abstract: The current mainstream semantic segmentation algorithms still have problems such as loss of small-sized objects and inaccurate segmentation. In response to these problems， this paper improves the HRNet network model and integrates the attention mechanism to generate more effective feature maps. To address the problem of insufficient detail of the feature map caused by the direct fusion of the low resolution images to the high-resolution images in the original model， a multi-level upsampling mechanism is proposed to make the fusion between images of different resolutions smoother to achieve better fusion results， and the depth separable convolution is used to reduce the parameters of the model. The model in this article maintains a high resolution of the image throughout the entire process. The spatial information of the feature map is improved， and the segmentation effect of small-sized objects is improved. The mIoU value on the PASCAL VOC2012 enhanced dataset reaches 80.87%， and the accuracy is improved by 1.54 percentage points compared with the original model.

A Weakened Joint Reinforcement Method to Improve Robustness of Image Recognition Models

LI Shi-da, XIANG Jian-wen

2023, 0(10): 70-76. doi:10.3969/j.issn.1006-2475.2023.10.011

Asbtract ( 49 )

PDF (2633KB) ( 54 )

References | Related Articles | Metrics

How to enhance the robustness of the model against adversarial examples attacks is an important research direction. In this paper， a method to improve the robustness of image recognition models is proposed. The method consists of a weakening operation and a strengthening operation. The weakening operation weakens the pixel values of the native input and destroys the structure of the adversarial perturbation. This process reduces the adversarial perturbation in the image but also loses some spatial semantic information， and this lost semantic information is supplemented by the reinforcement operation. The reinforcement operation consists of a feature extractor and a feature selector. The feature extractor is used to extract suitable image feature maps， and in order to select robust parts from these feature maps， a feature selector is designed to fuse the content of the feature maps and output feature maps with less perturbation and rich spatial semantic information. In this paper， the effectiveness of the method against adversarial examples is confirmed by extensive comparison experiments and the error accumulation phenomenon of adversarial perturbation is revealed.

A Fast Failure Recovery Scheme for SDN Based on Flow Aggregation and Congestion Avoidance

JIANG Hou-hai, ZHUANG Yi, CAO Zi-ning

2023, 0(10): 77-83. doi:10.3969/j.issn.1006-2475.2023.10.012

Asbtract ( 38 )

PDF (2000KB) ( 43 )

References | Related Articles | Metrics

Aiming at the problem that the proactive failure recovery scheme ignores the TCAM storage resource consumption of the backup path and the congestion after failure recovery， this paper proposes an SDN fast failure recovery scheme FACAR based on flow aggregation and congestion avoidance. FACAR is a proactive and fast failure recovery scheme with congestion-aware and low storage overhead. It considers all flows passing through the same link as one or several aggregate flows， and configures protection paths for these aggregate flows in advance. In this paper， FACAR is formalized as an integer linear programming problem， and a greedy-based heuristic algorithm ILP-FACAR is proposed to find the minimum number of configuration backup forwarding rules. Experimental results show that FACAR can meet the needs of the fast recovery after single link failure， and compared with other failure recovery methods， FACAR can ensure that there is no link congestion in the network after failure recovery， and greatly reduce the TCAM resource consumption of backup flow rules.

Real-time Behavioral Safety Warning for Power Operators Based on Multi-source Data

ZHANG Nan, LI Wen-jing, LIU Cai, XIE Ke, MA Shi-qian, XIAO Jun-hao, ZOU Feng

2023, 0(10): 84-91. doi:10.3969/j.issn.1006-2475.2023.10.013

Asbtract ( 90 )

PDF (2742KB) ( 56 )

References | Related Articles | Metrics

In order to reduce the occurrence of safety accidents and ensure the safety of power operators in the process of power grid construction， a behavior recognition model based on decision fusion of three dimensional residual convolutional neural network （R3D） models is proposed. First， the captured video dataset is subjected to data cleaning and enhancement； then， the corresponding R3D models are trained with the datasets collected from multiple angles； further， the multiple R3D models are fused at the decision level； finally， the possible violations or dangerous behaviors of power operators are warned in real-time by building a cloud platform. The experimental results show that the model has the advantages of high recognition accuracy and a low number of parameters， which proves that the behavior safety early warning method proposed in this paper can make early warning quickly and accurately and provide a safety guarantee for power grid construction.

Automatic Epistemic Emotion Recognition Based on Facial Expression in E-learning

CHEN Zi-jian, DUAN Chun-hong

2023, 0(10): 92-98. doi:10.3969/j.issn.1006-2475.2023.10.014

Asbtract ( 58 )

PDF (2219KB) ( 45 )

References | Related Articles | Metrics

The explicit facial expressions of learners provide a crucial measure for exploring their implicit epistemic emotions. A successfully accurate recognition of the epistemic emotion facial expressions in a real e-learning environment is still challenging due to its low change in intensity and short duration. In this paper， a new dual-modality spatiotemporal feature representation learning for recognizing facial expression in e-learning is proposed. Spatiotemporal geometrical feature representations and spatial-temporal appearance feature representations of facial expressions are designed to be automatically extracted with a hybrid deep neural network. The dual-modality feature fusion representations are used to recognize facial expressions. First， the experiment of micro-expression recognition is conducted on a spontaneous micro-expression dataset. The experimental result shows that the proposed method achieves higher recognition accuracy compared to the state-of-the-art methods. Next， a dataset of facial expression of epistemic emotions is created. Then， the recognition experiment of facial expression of epistemic emotions is conducted， and the model of micro-expression recognition is used in the model training of facial expression recognition of epistemic emotions by transfer learning. The multiple metrics are adopted to evaluate the model of facial expression recognition of epistemic emotions， and the experimental results demonstrate that the model is robust and efficient for the facial expression recognition of epistemic emotions.

Design of Power Material Sharing Cloud Warehouse Based on Blockchain Technology

WANG Guang-hui, CHENG Gong-xu, LI Qing

2023, 0(10): 99-106. doi:10.3969/j.issn.1006-2475.2023.10.015

Asbtract ( 54 )

PDF (2391KB) ( 46 )

References | Related Articles | Metrics

Aiming at the problems such as insufficient business information sharing and information asymmetry in the current power material supply chain， which cause complicated information flow scenarios， low efficiency and high supply costs， this paper constructs a power material supply chain sharing cloud warehouse based on blockchain technology. In the process of building a supply chain shared cloud warehouse， this article puts forward an idea of supply chain information sharing system， constructs an enterprise supply chain architecture， and realizes the sharing and circulation of equipment demand information of Internet power enterprises among multiple entities and multiple links. At the same time， in view of the high redundancy of blockchain storage， when building the supply chain information sharing system， a recording method of multi-source data information blocks inside and outside the supply chain is proposed to achieve the safe storage of data information， and a hierarchical access authority mechanism is built for each subject. Role-based access control is used to allocate the access authority of each enterprise to solve the problem of low transparency of access authority among supply chain subjects. Aiming at the problem that highly concurrent transaction uplink requests affect the blockchain networking performance， we monitor the throughput of the material sharing cloud warehouse blockchain nodes， propose a flow control algorithm， maximize the resource utilization of the underlying nodes of the blockchain， design an information smart contract， establish a supply chain information sharing system， and improve the reliability of the shared cloud warehouse. The experiment shows that the system has stable fault tolerance mechanism and performance， transparent sharing of business information， and high efficiency and reliability of sharing cloud warehouse when facing large-scale business volume.

Worm and Agent-based Attack Modeling for Industrial Control Systems

HAN Dong-song, SHA Le-tian, ZHAO Chuang-ye

2023, 0(10): 107-114. doi:10.3969/j.issn.1006-2475.2023.10.016

Asbtract ( 63 )

PDF (1971KB) ( 45 )

References | Related Articles | Metrics

In the field of network security， only by better understanding the attack， can we master the defense technology. This article focuses on the industrial control equipment in the industrial control system that is closest to the industrial production equipment - the programmable logic controller PLC， which is no longer limited to the traditional “host computer-PLC-cascading equipment” attack mode. Through the combination of PLC worm and PLC agent， the attack mode of “PLC-PLC-cascade device” with stronger attack adaptability is realized， and a complete attack chain that can make all PLCs in the Intranet environment be attacked by the PLC exposed to the directly accessible environment is realized. Different attack forms are added to the attack chain and the attack model is finally constructed. By building an experimental environment to conduct simulation experiments， it is proved that the attack model can change the operation state of the industrial control system and pose a threat to the safe operation of the industrial control system. Finally， targeted protection suggestions are given for this attack mode.

GAN-based Adversarial Attacks on Face Recognition

WANG Xin, XIAO Tao-rui

2023, 0(10): 115-120. doi:10.3969/j.issn.1006-2475.2023.10.017

Asbtract ( 95 )

PDF (1551KB) ( 54 )

References | Related Articles | Metrics

Face recognition is gradually becoming a monitoring tool which posed enormous threats to human privacy. For this reason， the paper proposes a semantic adversarial attack based on generative adversarial networks called SGAN-AA that modifies the significant facial features for images. It predicts the most significant attributes by using cosine similarity or probability score， and uses one or more facial features in white-box and black-box settings for impersonation and dodging attacks. The experimental results show that the method can generate diverse and realistic adversarial facial images while avoiding affecting human perception of facial recognition. The success rate of SGAN-AA's attack on black box models is 80.5%， which is 35.5 percentage points higher than common methods under impersonation attacks. Predicting the most significant attributes will improve the success rate of adversarial attacks in both white-box and black-box settings， and can enhance the transferability of the generated adversarial examples.

Two-factor Authentication Scheme Based on Smart Contract

LIU Xin, LIU Yi

2023, 0(10): 121-126. doi:10.3969/j.issn.1006-2475.2023.10.018

Asbtract ( 55 )

PDF (1287KB) ( 30 )

References | Related Articles | Metrics

Due to the rise of cryptocurrencies underpinned by blockchain and cryptography， the traditional centralized transaction model has been broken. But while it brings many conveniences， it also exposes flaws. When the user key of the cryptocurrency is lost or an attacker exploits the contract vulnerability to illegally transfer funds， the system lacks additional authentication and fund custody functions， which will cause the user to lose control of the funds. In view of these problems, the proposed proposal will write the user’s account fund tracking management rules into the smart contract, and force the user to call the two-factor authentication scheme （combined with non-interactive zero-knowledge proof， Merkle tree， ElGamal algorithm and other methods） to verify the legal identity and prevent attackers from illegally transferring funds under specific abnormal circumstances. The results of comparison with other schemes show that the scheme has a certain improvement in safety and efficiency.

Table of Content