MB-STFormer: A Multi-Band Spectral-Temporal Transformer with Efficient Attention for Enhanced EEG-Based Fatigue...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Transportation Safety

Background:

• Driver fatigue is a major cause of traffic accidents.
• Electroencephalogram (EEG) signals are reliable physiological indicators of fatigue.
• Existing methods struggle to capture the complex spatiotemporal-spectral dynamics of EEG signals.

Purpose of the Study:

• To develop a novel deep neural network, MB-STFormer, for accurate EEG-based driver fatigue detection.
• To integrate neurophysiological priors into deep feature learning for enhanced fatigue monitoring.
• To improve the interpretability and generalizability of fatigue detection systems.

Main Methods:

• Proposed MB-STFormer, a deep neural network integrating neurophysiological priors.
• Employed a multi-branch frequency-aware module to extract spatiotemporal features from EEG across different frequency sub-bands.
• Introduced an Efficient Additive Attention mechanism to aggregate global contextual information and mitigate feature over-smoothing.

Main Results:

• MB-STFormer achieved state-of-the-art performance in EEG-based fatigue detection.
• The model demonstrated superior interpretability and generalizability across three public datasets.
• The proposed framework effectively captures intricate spatiotemporal-spectral dynamics of EEG signals.

Conclusions:

• MB-STFormer offers a promising and accurate solution for real-world driver fatigue monitoring.
• The integration of neurophysiological priors enhances deep learning models for fatigue detection.
• The developed attention mechanism improves the capture of subtle EEG features, crucial for fatigue assessment.