Emotion recognition using hierarchical spatial-temporal learning transformer from regional to global brain.

Area of Science:

• Neuroscience
• Computer Science
• Artificial Intelligence

Background:

• Emotion recognition is crucial for human-computer interaction but challenging due to complex EEG signal patterns.
• Existing methods struggle to fully capture spatial and temporal dependencies of EEG signals across brain regions for accurate emotion recognition.

Purpose of the Study:

• To develop a transformer-based method, R2G-STLT, for enhanced emotion recognition from EEG signals.
• To effectively learn representative spatiotemporal features from electrode to brain-region levels.

Main Methods:

• Designed a spatial-temporal transformer encoder with regional to global hierarchical learning (R2G-STLT).
• Employed a regional spatial-temporal transformer (RST-Trans) for electrode-level feature extraction.
• Utilized a global spatial-temporal transformer (GST-Trans) with multi-head attention for brain-region level feature extraction.

Main Results:

• The R2G-STLT model demonstrated superior performance in subject-independent emotion recognition tasks.
• Evaluated on DEAP, SEED, and SEED-IV datasets across different frequency bands.
• Outperformed several state-of-the-art methods in capturing intricate spatiotemporal EEG features.

Conclusions:

• The proposed R2G-STLT method effectively captures hierarchical spatiotemporal features from EEG signals for emotion recognition.
• The regional to global learning approach enhances the model's ability to leverage information from different brain regions.
• R2G-STLT offers a promising advancement for emotion recognition in human-computer interaction systems.