DAVG-ViT: a domain-adaptive Vision Transformer with GA-BiLSTM for EEG emotion recognition

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Electroencephalography (EEG)-based emotion recognition is a growing field due to its non-invasive nature and high temporal resolution.
• Existing models face challenges in global dependency modeling, cross-channel interactions, and domain generalization.

Purpose of the Study:

• To propose a novel model, DAVG-ViT, for improved EEG-based emotion recognition.
• To address limitations in global modeling, cross-channel relationships, and domain adaptation.

Main Methods:

• The proposed DAVG-ViT model integrates Adapters for enhanced global modeling.
• Group Attention and BiLSTM are employed to capture cross-channel relationships and bidirectional temporal dynamics.
• Adversarial domain adaptation using a gradient reversal layer minimizes distribution shift.

Main Results:

• DAVG-ViT achieved superior performance on benchmark datasets: 99.37% accuracy on SEED, 95.91% on SEED-IV.
• On the DEAP dataset, accuracies of 98.95% for arousal and 98.47% for valence were obtained.

Conclusions:

• The DAVG-ViT model demonstrates significant advancements in EEG emotion recognition.
• The proposed methods effectively address challenges in global modeling, cross-channel interactions, and domain generalization.