基于自注意力机制编码的跨被试脑电情绪识别  

EEG Subject-independent Emotion Recognition Based on Self-attention Mechanism

作　　者：龚军辉[1] 袁超 田娟秀 GONG Junhui;YUAN Chao;TIAN Juanxiu(College of Electrical and Information Engineering,Hunan Institute of Engineering,Xiangtan 411104,China)

机构地区：[1]湖南工程学院电气与信息工程学院,湘潭411104

出　　处：《湖南工程学院学报(自然科学版)》2024年第4期38-43,共6页Journal of Hunan Institute of Engineering(Natural Science Edition)

基　　金：湖南省教育厅创新平台开发基金项目(20K036);湖南省自科基金面上资助项目(2021JJ30186).

摘　　要：脑电信号(electro encephalo gram,EEG)因具有时间分辨率高、采集无创等优点被广泛应用于情绪识别.由于个体大脑的差异性,跨被试脑电情绪识别性能的提高一直是个难题.针对上述问题及大脑信号的特性,提出了一种基于自注意力编码和密集连接卷积网络模型的跨被试脑电情绪识别方法.所提方法首先对脑电信号进行下采样、滤波和移除眼动成分等预处理;然后采用基于概率稀疏自注意力机制的编码层,学习不同脑区之间EEG信号的复杂关系;最后应用密集连接卷积网络获取时空域特征并进行分类识别.在公开的DEAP数据库上测试结果显示,本文所提方法情绪识别准确率达到64.8%,为跨被试EEG情绪识别提供了一种新方法.Electro encephalo gram(EEG)signals,which have advantages such as high temporal resolution and non-invasive acquisition,are widely used for emotion recognition.However,subject-independent emotion recognition of electro encephalo gram signals remains a challenge for individual differences.According to the characteristics of EEG signals,a subject-independent EEG-based emotion recognition method is proposed,which applied self-attention mechanism and Densely connected convolutional networks(DenseNet)to classify EEG signals in this paper.First,EEG signals are preprocessed by filtering,down sampling,and removing the artefacts.Then,the encoder based on probsparse self-attention is used to learn the complex relationships between EEG signals of different brain regions.Finally,DenseNet is applied to learn the spatiotemporal representation of EEG signals and then to classify EEG signals.Experiments on the publicly available DEAP database are conducted.The experimental results show that a relatively high average accuracy(64.8%)is obtained.The research provides a novel method for subject-independent emotion recognition.

关 键 词：脑电信号 情绪识别 自注意力机制 跨被试识别 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]