Tensor-Based EEG Network Formation and Feature Extraction for Cross-Session Driving Drowsiness Detection.

Area of Science:

• Neuroscience
• Traffic Safety
• Biomedical Engineering

Background:

• Drowsy driving is a significant global cause of traffic accidents.
• Existing electroencephalography (EEG) feature extraction methods for drowsiness detection often focus on individual channels, leading to variability issues.
• Robust detection systems are needed to mitigate risks associated with driver fatigue.

Purpose of the Study:

• To introduce a novel Tensor Network Features (TNF) method for enhanced drowsiness detection using EEG data.
• To address the limitations of existing methods in handling inter-session and inter-subject variability.
• To improve the accuracy and robustness of driving drowsiness detection systems.

Main Methods:

• Utilized Tucker decomposition on tensorized EEG channel data.
• Extracted features from subspace matrices using tensor network summation.
• Evaluated the TNF method on a sustained-attention driving task EEG dataset.

Main Results:

• The TNF method demonstrated improved accuracy compared to spectral power and fuzzy entropy features.
• Average accuracy improvements were 6.7% and 10.3% respectively.
• Maximum accuracy gains reached 17.3% and 29.7%.

Conclusions:

• The proposed Tensor Network Features (TNF) method offers a promising approach for practical and robust cross-session driving drowsiness detection.
• This method effectively exploits the underlying structure of drowsiness patterns in EEG data.
• Further development could lead to enhanced road safety systems.