动态图卷积联合记忆网络情绪脑电识别方法  

EEG-based emotion recognition using fusion model of graph convolutional neural networks and LSTM

作　　者：李浩 张学军[1,2] LI Hao;ZHANG Xuejun(College of Electronic and Optical Engineering,College of Flexible Electronics(Future Technology),Nanjing University of Posts and Telecommunications,Nanjing 210023,China;National and Local Joint Engineering Laboratory of RF Integration and Micro-Assembly Technology,Nanjing University of Posts and Telecommunications,Nanjing 210023,China)

机构地区：[1]南京邮电大学电子与光学工程学院、柔性电子(未来技术)学院,南京210023 [2]南京邮电大学射频集成与微组装技术国家地方联合工程实验室,南京210023

出　　处：《智能计算机与应用》2025年第1期203-210,共8页Intelligent Computer and Applications

基　　金：国家自然科学基金(61977039)。

摘　　要：针对无法有效利用脑电通道拓扑结构学习更有鉴别性的脑电特征问题,本文基于长短期记忆网络和图卷积神经网络,提出动态图卷积联合记忆网络(Dynamic Graph Convolutional Joint Long Short Term Memory Network,DGCJMN)方法。首先将脑电通道作为图的节点,微分熵作为节点特征,利用动态参数学习最优的脑电通道拓扑结构,构建特征图;之后,由图卷积神经网络提取图域特征,并结合长短期记忆网络和池化进一步提取特征;最后将图卷积网络、长短期记忆网络和池化提取的特征融合后进行情绪分类。所提方法在SEED数据集上针对积极、中性和消极3种情绪取得的平均准确率为95.93%,精确率、召回率和F1值分别为96.11%、95.93%和0.96,Kappa系数为0.939。混淆矩阵表明,模型对于3种情绪都达到了较好的分类效果。Aiming at the problem that the topology structure of EEG channels cannot be effectively used to learn more discriminative EEG features,this paper proposes the dynamic graph convolutional joint memory network(Dynamic Graph Convolutional Joint Long Short Term Memory Network,DGCJMN) based on the long and short term memory network and the graph convolutional neural network.First,the EEG channel is taken as the node of the graph,and the differential entropy is taken as the node feature,and the optimal EEG channel topology is learned by dynamic parameters to construct the feature graph.After that,the features of the graph domain are extracted by the convolutional neural network,and further extracted by combining the long short-term memory network and pooling.Finally,the features extracted by Graph Convolutional Network,Long Short-term Memory network and pooling were fused for emotion classification.The proposed method achieved an average accuracy of 95.93% for positive,neutral and negative emotions in SEED dataset;accuracy,recall and F1 scores were 96.11%,95.93% and 0.96;Kappa coefficient was 0.939;confusion matrix indicated that the model achieved a good classification effect for the three emotions.

关 键 词：情绪识别 脑电图 图卷积神经网络 长短期记忆网络 微分熵 

分 类 号：TP274[自动化与计算机技术—检测技术与自动化装置] R318[自动化与计算机技术—控制科学与工程]