SATEER: Subject-Aware Transformer for EEG-Based Emotion Recognition.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Computer Vision

Background:

• Electroencephalogram (EEG) signals offer insights into human emotional states.
• Accurate emotion recognition from EEG is challenging due to individual variability.
• Existing methods often overlook personalized responses to stimuli.

Purpose of the Study:

• To develop a Subject-Aware Transformer-based neural network (SATEER) for enhanced EEG emotion recognition.
• To address individual response variability by incorporating a User Embedder module.
• To improve the accuracy and robustness of emotion classification from EEG data.

Main Methods:

• EEG waveforms were transformed into Mel spectrograms, enabling processing via a Computer Vision pipeline.
• A Subject-Aware Transformer architecture was employed, incorporating a User Embedder module.
• The model was evaluated on four publicly available EEG emotion recognition datasets.

Main Results:

• SATEER demonstrated superior performance across all benchmarked datasets compared to existing methods.
• On the AMIGOS dataset, SATEER achieved over 99.8% accuracy, surpassing the state-of-the-art by 0.47%.
• Ablation studies confirmed the critical contribution of the User Embedder module and other model components.

Conclusions:

• The proposed SATEER model significantly advances EEG-based emotion recognition.
• The User Embedder module is crucial for handling individual differences in EEG emotion analysis.
• SATEER offers a robust and highly accurate solution for classifying human emotional states from EEG signals.