CNN-RxLSTM:基于混合卷积和残差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(EEG)  CNN  扩展型长短时记忆网络(xLSTM)  残差机制

Key words:  Emotion Recognition  Electroencephalogram (EEG)  CNN  Extended Long Short-Term Memory Network (xLSTM)  Residual Mechanism

摘要:

针对基于脑电图(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,提出了一种基于混合时空卷积CNN和残差的扩展型长短时记忆网络(简称残差xLSTM)的EEG情绪识别网络CNN-RxLSTM。该网络框架使用CNN提取EEG信号的局部时空特征,并引入了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,有效提升了在情绪识别中的准确率。在SEED数据集上实现了98.15%的分类准确率,在DEAP数据集中,效价(valence)和唤醒(arousal)的分类准确率分别达到94.60%和95.89%。结果充分验证了该模型在EEG情绪识别中的卓越性能,同时为情绪解码和其他EEG相关应用提供了新的解决方案。

Abstract:

The EEG-based emotion recognition method faces challenges in feature extraction and modeling spatiotemporal dependencies. To address these limitations, a new EEG emotion recognition network, CNN-RxLSTM, based on hybrid spatiotemporal convolutional CNN and residual Extended Long Short-Term Memory (RxLSTM), was proposed. The network framework used CNN to extract local spatiotemporal features of EEG signals and employed the xLSTM module to model global spatiotemporal dependencies through bidirectional information flow processing and residual connection mechanisms. This approach effectively improved the accuracy of emotion recognition. The model achieved a classification accuracy of 98.15% on the SEED dataset, and classification accuracies of 94.60% for valence and 95.89% for arousal on the DEAP dataset. The results demonstrate the superior performance of the CNN-RxLSTM framework in EEG emotion recognition and provide a new solution for emotion decoding and other EEG-related applications.