سامانه همایش‌ها TSTA | ثبت‌نام و ارسال مقاله کنفرانس‌ها و کنگره‌های ملی

صفحه اصلی / سی و دومین کنفرانس ملی و دهمین کنفرانس بین المللی مهندسی زیست پزشکی ایران

Vision Transformer-Based Emotion Recognition in EEG Using Pseudo-Image Construction

نویسندگان :

Ali Kouchakzadeh¹ Soheil Moradi² Mohammad Mohsen Ebrahimi Seyghalan³ Mehdi Fardmanesh⁴

1- دانشگاه صنعتی شریف، تهران، ایران 2- دانشگاه صنعتی شریف، تهران، ایران 3- دانشگاه صنعتی شریف، تهران، ایران 4- دانشگاه صنعتی شریف، تهران، ایران

کلمات کلیدی :

EEG data،transformers،emotion recognition

چکیده :

Transformer-based models have demonstrated remarkable performance across natural language processing and computer vision. Our research aims to extend these models to time-series EEG data, evaluating their effectiveness for feature extraction and pattern recognition. To achieve this, we first transform the EEG data into pseudo-images using autoencoders. This approach enables us to convert multi-channel time-series data into spatio-temporal 2D pseudo-images, which are then processed by a Vision Transformer (ViT) model. Two important features in EEG data that we need to consider are intra-channel features, which represent the dynamics within each individual channel, and inter-channel features, which capture the relationships between different channels. In our method, the autoencoder captures the unique intra-channel features of each EEG channel, while the self-attention mechanism in the ViT uncovers inter-channel dependencies, enhancing the model’s representational power. Additionally, this approach provides visualizations of active channels during different emotional states, offering insights into brain function and supporting further research in understanding the neural basis of emotions. We evaluated our model on two widely recognized EEG datasets, SEED and DREAMER, achieving promising results: 99.81% accuracy on SEED, and 97.17% for valence and 97.50% for arousal on DREAMER, outperforming existing models. These findings suggest that the proposed method effectively extracts rich and meaningful features from EEG data, paving the way for more advanced analysis in neural and affective computing.