A recurrence network-based convolutional neural network for fatigue driving detection from EEG.

Area of Science:

• Neuroscience
• Computer Science
• Traffic Safety Engineering

Background:

• Driver fatigue is a significant contributor to traffic accidents.
• Existing fatigue detection methods include traditional feature-based and machine learning approaches.
• Deep learning techniques have shown promise, with hybrid methods gaining attention.

Purpose of the Study:

• To propose a novel Recurrence Network-based Convolutional Neural Network (RN-CNN) for detecting driver fatigue.
• To evaluate the effectiveness of the proposed RN-CNN method using electroencephalogram (EEG) signals.

Main Methods:

• A simulated driving experiment was conducted to collect EEG data from alert and fatigued drivers.
• A multiplex recurrence network (RN) was constructed from EEG signals to fuse time-series information.
• A Convolutional Neural Network (CNN) was utilized to extract features from the multiplex RN for classification.

Main Results:

• The proposed RN-CNN method achieved an average accuracy of 92.95% in detecting driver fatigue.
• Comparative analysis demonstrated that the RN-CNN method outperforms existing competitive methods.
• The results validate the efficacy of the RN-CNN approach for fatigue detection.

Conclusions:

• The developed RN-CNN method offers a highly accurate and effective solution for driver fatigue detection.
• This approach holds potential for enhancing road safety by mitigating fatigue-related accidents.