STAFNet: an adaptive multi-feature learning network via spatiotemporal fusion for EEG-based emotion recognition

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Emotion recognition via electroencephalography (EEG) is crucial for brain-computer interfaces.
• Current methods often neglect the integration of spatial and temporal features, limiting performance.
• Effective emotion recognition demands capturing complex brain dynamics.

Purpose of the Study:

• To develop a novel framework, Spatiotemporal Adaptive Fusion Network (STAFNet), for accurate and robust EEG-based emotion recognition.
• To enhance the integration of spatial and temporal features for improved emotion classification.
• To address limitations in existing EEG emotion recognition techniques.

Main Methods:

• Proposed the Spatiotemporal Adaptive Fusion Network (STAFNet).
• Utilized adaptive graph convolution for spatial dynamic evolution and brain connectivity patterns.
• Employed a multi-structured transformer fusion module to integrate spatial and temporal features.

Main Results:

• Achieved high accuracies of 97.89% on the SEED dataset and 93.64% on the SEED-IV dataset.
• Outperformed existing state-of-the-art methods in EEG-based emotion recognition.
• Demonstrated STAFNet's ability to mitigate over-smoothing issues in deep graph convolutional networks.

Conclusions:

• Validated the effectiveness of STAFNet for EEG-based emotion recognition.
• Highlighted the critical role of spatiotemporal feature extraction and fusion.
• Advanced the state of the art in emotion recognition through an innovative fusion framework.