Abstract
Fatigue driving detection is essential to prevent traffic accidents and ensure driving safety. Electroencephalogram (EEG) signals serving as the key physiological indicators of brain activity are widely used to assess the driving fatigue. However, the main challenge in EEG-based fatigue detection is the efficient extraction of features closely related to driving fatigue. This study explores preprocessing and feature extraction methods for fatigue EEG signals. Addressed the susceptibility of raw EEG signals to electrooculography (EOG) artifact interference, an innovative method combining Ensemble Empirical Mode Decomposition (EEMD) and Fast Independent Component Analysis (FastICA) is proposed. This method can effectively filter out EOG artifacts, resulting in purer EEG signals. Additionally, to overcome the limitations of single-method feature extraction and enhance detection accuracy, a novel strategy integrating Wavelet Packet Transform (WPT) and Sample Entropy (SampEn) is introduced. This approach first employs WPT to extract time-frequency features from purer EEG signals, then applies SampEn to capture their nonlinear features, and finally integrates these features into a comprehensive vector, which is classified using an Support Vector Machine (SVM). The experimental results demonstrate that compared to the single feature extraction methods, the proposed multi-feature fusion approach can more effectively capture the detailed information in fatigue EEG signals, significantly improving the recognition accuracy of driving fatigue detection.