Enhancing Emotion-Brain Representations With Orthogonal Fuzzy Power-Coherence Alignment

Electroencephalogram (EEG)-based emotion recognition is a cornerstone of affective computing and human-computer interaction. However, progress in this field remains limited by the inherent fuzziness of emotions and the high complexity of EEG signals. To address these challenges, we propose Orthogonal Fuzzy Power-Coherence Alignment (OFPCA), a novel framework designed to enhance emotional brain representation learning. To effectively model the fuzzy nature of emotions, OFPCA employs a Takagi-Sugeno-Kang (TSK) fuzzy system as a feature learner. To overcome the complexity of EEG signals, it aligns two complementary perspectives-the local-energy view and the cross-regional interaction view-through contrastive learning, enabling a more comprehensive EEG representation. The local-energy view is represented by power spectral density (PSD), while the cross-regional interaction view is captured by coherence (COH). To ensure robust emotional brain representation learning, we introduce orthogonal regularization loss into OFPCA's training process. We evaluate OFPCA on two benchmark datasets (SEED with three emotions and CRED with five emotions) under two evaluation settings. Under the cross-subject setting, OFPCA achieves 41.14% on CRED and 83.24% on SEED. Under the intra-subject setting, it reaches 63.24% on CRED and 85.37% on SEED. These results surpass existing methods and demonstrate the effectiveness of OFPCA. This study demonstrates that leveraging the inherent fuzziness of emotions enhances emotion recognition and introduces OFPCA as a novel fuzzy learning-based framework for multi-view EEG learning.