A fatigue driving detection method based on local maximum refined composite multi-scale normalized dispersion entropy and SVM.

Multi-scale dispersion entropy (MDE) has been extensively applied to capture the nonlinear features of electroencephalography (EEG) signals for fatigue driving detection. However, MDE suffers from information loss and limited robustness during the extraction of EEG signal nonlinearities. To address these issues, a fatigue driving detection approach integrating local maximum refined composite multi-scale normalized dispersion entropy (LMRCMNDE) with support vector machines (SVM) is introduced. To begin, the refined composite multi-scale dispersion entropy (RCMDE) technique is presented. Next, the segmented averaging in the coarse-graining process is substituted with local maximum calculation to alleviate information loss. Finally, normalization of the entropy values is performed to enhance the robustness of feature parameters, leading to the formation of LMRCMNDE. LMRCMNDE serves as the feature descriptor for fatigue driving EEG signals, while SVM is employed for classification. Compared with the MDE-SVM and RCMDE-SVM approaches, the LMRCMNDE-SVM method achieves higher recognition accuracy, reaching up to 98%. The proposed method can effectively identify the fatigue state of drivers and provide a new reliable detection method for automatic fatigue driving detection.