Deep Learning-Based Emotion Recognition from EEG Signals: A Hybrid Architecture with EEGNet and Transformer Derin Öǧrenme ile EEG Sinyallerinden Duygu Tanima: EEGNet ve Transformer Tabanli Hibrit Mimari
34th Signal Processing and Communications Applications Conference, SIU 2026, İstanbul, Türkiye, 7 - 10 Temmuz 2026, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/siu71813.2026.11636536
- Basıldığı Şehir: İstanbul
- Basıldığı Ülke: Türkiye
- Anahtar Kelimeler: deep learning, EEG, EEGNet, emotion recognition, SEED dataset, Transfor-mer
- Gazi Üniversitesi Adresli: Evet
Özet
Automated recognition of human emotions has be-come a significant research area in fields such as brain-computer interfaces, human-computer interaction, and affective computing. Emotion recognition from electroencephalography (EEG) signals is considered a vital approach as it provides direct access to emotional states through brain activity. However, the high dimensionality, low signal-to-noise ratio, and complex spatio-temporal structure of EEG data pose challenges for developing effective and generalizable models. In this study, we propose a hybrid deep learning architecture for the EEG-based emotion classification problem that integrates domain-specific information directly into the model. Using the widely recognized SEED dataset, raw EEG signals are processed through end-to-end learnable layers instead of predefined feature extraction methods. The proposed architecture combines an EEGNet-based feature extraction block with a Transformer encoder. While EEGNet generates representations suitable for the temporal and spatial structure of EEG signals, the Transformer models long-term temporal dependencies using its multi-head attention mechanism. Experimental results demonstrate that the proposed EEGNet + Transformer approach achieves an accuracy of approximately 96.31% on the SEED dataset, offering a competitive performance compared to state-of-the-art methods in the literature.