GCAN-DA: Emotion EEG Recognition with Graph Convolutional Attention and Domain Adaptation

Electroencephalography (EEG) measures neuronal activity across brain regions, but many existing emotion recognition studies do not fully exploit the topological structure of EEG channels. In this paper, we propose a Graph Convolutional Attention Network with Domain Adaptation (GCAN-DA), which models inter-channel relationships using adjacency matrices inspired by brain neuroscience. The model includes three graph branches: local, global, and fusion, to automatically learn emotion-related brain patterns. Additionally, an attention mechanism is implemented to further refines the brain activation pat-terns linked to emotional cognition. The proposed model is evaluated based on the SEED dataset, achieving 87.33% accuracy in subject-independent experiments. These results demonstrate the effectiveness of GCAN-DA in EEG-based emotion recognition, offering a promising approach for emotional computing.