GLWGAN: Dual-Discriminator GAN for EEG Emotion Recognition Using Spatiotemporal Patterns

Emotion recognition frequently relies on electroencephalogram (EEG) signals, but their data scarcity and complexity pose challenges for model training. This paper introduces GLWGAN, a comprehensive neural network architecture designed for emotion recognition, utilizing a dual-discriminator GAN model to enhance data augmentation. This approach integrates global and local information to enhance EEG feature representation. The feature extraction and classification processes are subsequently performed using a Bidirectional Long Short-Term Memory (BiLSTM) network architecture. Comprehensive evaluations conducted on the MAHNOB-HCI and SEED datasets highlight the framework's strong performance, with average accuracy rates of 79.85% and 81.12% in predicting valence and arousal, respectively, and 88.69% in three-class emotion categorization on the SEED dataset. Our results indicate that the proposed framework significantly improves spatiotemporal information extraction and emotion recognition accuracy.