Self-supervised group meiosis contrastive learning for EEG-based emotion recognition
Haoning Kan,Jiale Yu,Jiajin Huang,Zihe Liu,Heqian Wang,Haiyan Zhou,Yu, Jiale
DOI: https://doi.org/10.1007/s10489-023-04971-0
IF: 5.3
2023-09-06
Applied Intelligence
Abstract:The progress of EEG-based emotion recognition has received widespread attention from the fields of human-machine interaction and cognitive science. However, recognizing emotions with limited labelled data is still challenging. To address this issue, this paper proposes a self-supervised group meiosis contrastive learning (SGMC) framework for EEG-based emotion recognition. First, to reduce the dependence of emotion labels, SGMC introduces a contrastive learning task according to the alignment of video clips based on the similar EEG response across subjects. Moreover, the model adopts a group projector to extract group-level representations from the group samples to further decrease the subject difference and random effects in EEG signals. Finally, a novel genetics-inspired data augmentation method, named meiosis is developed, which takes advantage of the alignment of video clips among a group of EEG samples to generate augmented groups by pairing, cross exchanging, and separating. The experiments show that SGMC exhibits competitive performance on the publicly available DEAP and SEED datasets. It is worth of noting that the SGMC shows a high ability to recognize emotion even when using limited labelled data. Moreover, the results of feature visualization suggest that the model might have learned video-level emotion-related feature representations to improve emotion recognition. The hyper-parametric analysis further shows the effect of the group size during emotion recognition. Finally, the comparisons of both the symmetric function and the ablation models and the analysis of computational efficiency are carried out to examine the rationality of the SGMC architecture. The code is provided publicly online.
computer science, artificial intelligence