EmT: A Novel Transformer for Generalized Cross-Subject EEG Emotion Recognition

Area of Science:

• Neuroscience
• Artificial Intelligence
• Signal Processing

Background:

• Emotion decoding from electroencephalography (EEG) benefits from integrating neurophysiology into neural network architectures.
• Current methods often overlook long-term contextual information crucial for emotional cognitive processes.

Purpose of the Study:

• Introduce a novel transformer model, the emotion transformer (EmT), for enhanced EEG-based emotion decoding.
• Improve generalized cross-subject emotion classification and regression tasks by capturing long-term temporal dependencies.

Main Methods:

• EEG signals are converted into a temporal graph format using a temporal graph construction (TGC) module.
• A residual multiview pyramid graph convolutional neural network (RMPG) learns dynamic graph representations.
• A temporal contextual transformer (TCT) module with specialized token mixers captures temporal context.

Main Results:

• EmT demonstrates superior performance in both EEG emotion classification and regression tasks across four public datasets.
• The proposed model effectively learns from spatial and short-term temporal patterns while emphasizing long-term context.

Conclusions:

• The EmT model offers a significant advancement in EEG-based emotion recognition by incorporating long-term contextual information.
• This approach holds promise for more accurate and robust emotion decoding in various applications.