K-means++ guided multi-view CNN with channel attention for EEG emotion recognition

The use of artificial intelligence for emotion recognition is the focus of improving human-computer interaction. Recently, deep learning has been widely used in the study of emotion recognition. However, how to correctly identify emotions still faces a huge challenge. We propose a multi-view deep CNN based on channel attention (MVACNN) for EEG emotion recognition. MVACNN first clustered the channels and divided the channels with high similarity into the same view. Channels within the same view have highly similar patterns of neural activity, which can extract the synergistic features of specific brain regions more intensively and reduce the interference of irrelevant noise. To extract more discriminative features, MVACNN integrates channel attention into each view. Channel attention gives different channel weights to effectively learn the importance of different channels. In addition, MVACNN uses residual blocks to learn the residual variation to better represent the relationship between input and output. The residual structure preserves the original information, improving feature utilization and thus maintaining performance. Experimental results show that MVACNN achieves good results on different datasets.