Two-Phase Prototypical Contrastive Domain Generalization for Cross-Subject EEG-Based Emotion Recognition

Abstract

EEG signals of different individuals belong to different domains and have different data distributions because great individual distinctions exist in EEG signals. The existing methods on EEG-based emotion recognition often ignore this property and need to collect extensive EEG data for new subjects to calibrate a brain-computer interface (BCI). In this paper, a two-phase prototypical contrastive domain generalization framework (PCDG) is proposed for cross-subject EEG-based emotion recognition, which mainly consists of a new convolutional neural network based on a residual block and a CBAM block and a two-phase prototypical representation-based contrastive learning method. The effectiveness of the proposed PCDG was evaluated on two public datasets (SEED and SEED-IV) with SVM and other baseline domain adaptation (DA) and domain generalization (DG) methods. The experimental results showed that the PCDG outperformed other baseline methods but without accessing the data of target domains.