ST-GCN: EEG Emotion Recognition via Spectral Graph and Temporal Analysis with Graph Convolutional Networks

Emotion recognition from electroencephalogram (EEG) signals is a key application in brain-computer interfaces (BCIs), but the high dimensionality and noise in EEG data pose significant challenges. Many existing approaches fail to adequately filter irrelevant information or fully capture complex inter-channel relationships and temporal dynamics, leading to suboptimal emotional representation.To address these challenges, we propose ST-GCN, a novel model that integrates spectral and temporal domain features using graph convolution for robust EEG emotion recognition. ST-GCN employs a channel information reconstruction layer, channel aggregation, and temporal feature extraction to learn discriminative representations across EEG channels and time. Evaluated on the DEAP dataset with a cross-validation setup, ST-GCN achieves state-of-the-art performance, with 98.43% accuracy for valence and 98.69% for arousal, demonstrating its effectiveness in EEG-based emotion recognition.