EEG Emotion Recognition Based on Dynamic Graph Neural Networks

Electroencephalography signals are increasingly used in affective computing. Taking into account subject differences, we propose a novel Graph Transformer network model based on self-attention and channel attention. The model addresses variations in EEG channel connectivity relationships among subjects by utilizing a mapping layer, while leveraging a Transformer-based graph neural network and channel attention to learn and enhance the representational capacities of the output. To validate the effectiveness of the proposed method, we conducted experiments on each of the 15 subjects of the SEED dataset. The average accuracy and standard deviation of the proposed method are 94.35% and 5.05%, respectively. The results show that our method outperforms the existing methods.