Driver fatigue recognition using limited amount of individual electroencephalogram

Area of Science:

• Neuroscience
• Biomedical Engineering
• Transportation Safety

Background:

• Driver fatigue is a major cause of road accidents.
• Existing fatigue detection methods often rely on physical cues or vehicle data, which can be unreliable.
• Electroencephalogram (EEG) signals offer a direct measure of the brain's physiological and mental state.

Purpose of the Study:

• To develop an accurate fatigue recognition system using EEG signals.
• To identify critical brain regions and frequencies for fatigue detection.
• To enhance system performance through data augmentation and transfer learning.

Main Methods:

• Utilized electroencephalogram (EEG) signals for fatigue state recognition.
• Applied transfer learning to partial ensemble averaged EEG power spectral density (PSD).
• Employed layer-wise relevance propagation (LRP) for identifying key cortical regions and frequency bands.
• Incorporated data augmentation techniques to improve classification accuracy.

Main Results:

• Achieved high classification accuracies: 99.2% (training), 97.9% (validation), and 96.9% (test).
• Demonstrated significant performance improvement with data augmentation.
• Identified critical cortical regions and frequency bands for effective fatigue discrimination.

Conclusions:

• EEG-based fatigue monitoring systems can significantly enhance road safety.
• Data augmentation is a beneficial technique for improving the accuracy of EEG fatigue detection.
• Further research should focus on optimizing data augmentation and system scalability.