CNN-RxLSTM:基于混合时空卷积和残差xLSTM的EEG情绪识别网络

CNN-RxLSTM: EEG Emotion Recognition Network Based on Hybrid Convolution and Residual xLSTM

关键词:  情绪识别  脑电图  卷积神经网络  扩展型长短时记忆网络  残差机制

Key words:  emotion recognition  Electroencephalogram (EEG)  Convolutional Neural Network (CNN)  Extended Long Short-Term Memory Network (xLSTM)  residual mechanism

摘要:

针对基于脑电图(Electroencephalogram,EEG)的情绪识别方法在特征提取不足和时空依赖建模的局限性，本研究提出一种基于混合时空卷积神经网络(Convolutional Neural Network,CNN)和残差扩展型长短时记忆网络(Residual Extended Long Short-Term Memory，RxLSTM)的EEG情绪识别模型CNN-RxLSTM。该模型首先使用CNN提取EEG信号的局部时空特征，然后引入xLSTM模块通过双向信息流处理和残差连接机制建模信号的全局时空依赖，最后通过分类器模块完成分类。为验证模型的有效性，分别在SEED数据集和DEAP数据集上进行实验。结果表明，在SEED数据集上，CNN-RxLSTM模型的分类准确率为98.15%；在DEAP数据集上，其效价分类准确率和唤醒分类准确率分别为94.60%和95.89%。研究结果验证了该模型在EEG情绪识别中的卓越性能，可为情绪解码和其他EEG相关研究提供新的解决方案。

Abstract:

Aiming at the limitations of Electroence-phalogram (EEG)-based emotion recognition methods in insufficient feature extraction and spatio-temporal dependence modeling,this study proposes an EEG emotion recognition network CNN-RxLSTM based on hybrid spatio-temporal Convolutional Neural Network (CNN) and Residual Extended Long Short-Term Memory (RxLSTM).The model first uses CNN to extract the local spatio-temporal features of EEG signals,then introduces the xLSTM module to model the global spatio-temporal dependence of signals through bidirectional information flow processing and residual connection mechanism.Finally,the classification is completed by the classifier module.In order to verify the validity of the model,experiments are carried out on SEED dataset and DEAP dataset respectively.The results show that the classification accuracy of the CNN-RxLSTM model is 98.15% on the SEED dataset.On the DEAP dataset,the accuracy of valence classification and arousal classification is 94.60% and 95.89%,respectively.The research results verify the excellent performance of the model in EEG emotion recognition,which can provide new solutions for emotion decoding and other EEG related research.