MBRSTCformer: a knowledge embedded local-global spatiotemporal transformer for emotion recognition.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Emotion recognition is crucial for advanced Brain-Computer Interfaces (BCIs).
• Electroencephalography (EEG) is widely used for real-time emotion mapping.
• Existing deep learning models often overlook crucial local brain region features in EEG signals.

Purpose of the Study:

• To propose a novel framework, MBRSTCfromer, for robust EEG-based emotion recognition.
• To address limitations of holistic EEG feature extraction by incorporating multi-brain region analysis.
• To enhance the accuracy and effectiveness of emotion recognition models for BCIs.

Main Methods:

• Developed the Multi-Brain Regions Collaboration Network to process EEG data segmented by brain regions.
• Introduced stimulation scores to quantify and feedback brain region activity.
• Proposed a Cascade Pyramid Spatial Fusion Temporal Convolution Network for integrating multi-brain region EEG features.

Main Results:

• Achieved 98.63% accuracy for arousal, 98.15% for valence, and 98.58% for dominance on the DEAP dataset.
• Attained 97.66% accuracy for arousal, 97.07% for valence, and 97.97% for dominance on the DREAMER dataset.
• Demonstrated the effectiveness of the MBRSTCfromer framework in comprehensive experiments.

Conclusions:

• The MBRSTCfromer framework significantly improves EEG-based emotion recognition.
• Collaborative analysis of multi-brain regions enhances feature extraction and model performance.
• The proposed model offers a robust solution for real-world BCI applications.