Dynamic domain adaptive EEG emotion recognition based on multi-source selection

Emotion recognition based on electroencephalogram (EEG) has always been a research hotspot. However, due to significant individual variations in EEG signals, cross-subject emotion recognition based on EEG remains a challenging issue to address. In this article, we propose a dynamic domain-adaptive EEG emotion recognition method based on multi-source selection. The method considers each subject as a separate domain, filters suitable source domains from multiple subjects by assessing their resemblance, then further extracts the common and domain-specific features of the source and target domains, and then employs dynamic domain adaptation to mitigate inter-domain discrepancies. Global domain differences and local subdomain differences are also considered, and a dynamic factor is added so that the model training process first focuses on global distribution differences and gradually switches to local subdomain distributions. We conducted cross-subject and cross-session experiments on the SEED and SEED-IV datasets, respectively, and the cross-subject accuracies were 89.76% and 65.28%; the cross-session experiments were 91.63% and 67.83%. The experimental outcomes affirm the efficacy of the EEG emotion recognition approach put forward in this paper.