Benchmarking EEG-based Cross-dataset Driver Drowsiness Recognition with Deep Transfer Learning.

Area of Science:

• Neuroscience
• Machine Learning
• Transportation Safety

Background:

• Electroencephalography (EEG) is crucial for driver drowsiness monitoring.
• Collecting calibration data for EEG is time-intensive.
• Cross-dataset recognition is needed to reduce calibration time but faces distribution drift challenges.

Purpose of the Study:

• To develop a deep transfer learning model for effective cross-dataset driver drowsiness recognition.
• To address the distribution drift problem in EEG datasets from different environments.
• To enable calibration-free driver drowsiness detection.

Main Methods:

• Proposed the Entropy-Driven Joint Adaptation Network (EDJAN), a deep transfer learning model.
• Utilized an entropy-driven loss function for representation clustering.
• Implemented an individual-level domain adaptation technique to mitigate distribution discrepancies.

Main Results:

• Achieved 83.3% accuracy with SADT as the source and SEED-VIG as the target domain.
• Obtained 76.7% accuracy with SEED-VIG as the source and SADT as the target domain.
• Outperformed state-of-the-art (SOTA) methods in cross-dataset recognition tasks.

Conclusions:

• The EDJAN model effectively handles distribution drift in cross-dataset EEG analysis.
• The proposed methods enable accurate, calibration-free driver drowsiness recognition.
• This research opens avenues for practical EEG applications in driver safety.