Emotion Recognition from EEG Signal Using CA-GCN

Abstract

Electroencephalography(EEG)-based emotion recognition has become an important task in the field of brain computer interface(BCI). Recently, deep learning methods have achieved excellent results in the field of EEG emotion recognition. However, these methods either ignore the relations between channels or ignore the contributions of different channels, which makes it difficult to improve the recognition performance. In this paper, an end-to-end EEG emotion recognition method that graph convolutional neural network with channel-wise attention(CA-GCN) is proposed. First, we utilize a GCN layer to extract EEG signals spatial information for expressing intrinsic relations between different EEG channels. Furthermore, a channel-wise attention pooling layer in CA-GCN was adopted to extract the contribution of different EEG channels. Sufficient experiments have been carried out on the DEAP database. Experimental results demonstrate that the proposed method outperforms the previous methods, and the recognition accuracy of the proposed method achieve 93.69% and 94.59% in valence and arousal, respectively.