Cross-subject emotion recognition with contrastive learning based on EEG signal correlations

Mengting Hu, Dan Xu, Kangjian He, Kunyuan Zhao, Hao Zhang

Abstract: Highlights•Powerful EEG Emotion Recognition Framework: CL-CS integrates multi-dimensional EEG features (channel, time, frequency) to enhance emotion classification across subjects.•Adapted Loss Function for EEG Signals: CL-CS introduces a contrastive loss based on signal correlations, improving model robustness by considering both amplitude and phase information.•Reduction of Cross-Subject Data Discrepancies: The 3-domain encoder trained via contrastive learning effectively reduces data discrepancies between different domains, enhancing emotion recognition generalization without large annotated datasets.•Improved Utilization of Limited Data: By combining contrastive learning with joint learning methods, CL-CS optimizes the use of available EEG data, improving classification performance and generalization ability for cross-disciplinary emotion recognition tasks.