EEG-based fatigue state evaluation by combining complex network and frequency-spatial features.

Area of Science:

• Neuroscience
• Machine Learning
• Transportation Safety

Background:

• Increasing rates of traffic accidents attributed to driver fatigue highlight the urgent need for effective detection methods.
• Current driver fatigue detection systems require improvement in speed and accuracy.

Purpose of the Study:

• To develop and validate a novel electroencephalogram (EEG)-based method for evaluating driver fatigue states.
• To combine complex network and frequency-spatial features for enhanced fatigue detection accuracy.

Main Methods:

• Construction of a complex network model using relative wavelet entropy to analyze inter-channel correlations.
• Extraction of frequency and spatial features through differential entropy and symmetry quotient calculations.
• Fusion of complex network and frequency-spatial features into a brain heat map.
• Application of a convolutional neural network-long short-term memory (CNN-LSTM) model for classifying three fatigue states (awake, tired, drowsy).

Main Results:

• The proposed method achieved an average classification accuracy of 96.57% on the SEED-VIG dataset for distinguishing between awake, tired, and drowsy states.
• On an external dataset from Mendeley Data, the method demonstrated an even higher average classification accuracy of 99.23%.
• The method outperformed existing state-of-the-art approaches in recognizing three-class fatigue states.

Conclusions:

• The study validates the effectiveness of an EEG-based fatigue evaluation method utilizing inter-channel correlations and frequency-spatial features.
• This approach holds significant promise for the development of advanced driver fatigue detection systems.
• The findings contribute to improving road safety by providing a reliable tool for monitoring driver alertness.