EEG-Based Driving Fatigue Detection in Urban Road Conditions Using a Novel Semi-Dry Electrode

To address the issue that conductive liquid in traditional electroencephalogram (EEG) electrodes cannot be replenished in time during long-term use and that the replenishment process is complex, this study proposes a novel semi-dry electrode with an automatic replenishment of conductive liquid function. Additionally, in the context of driving fatigue detection in urban road environments, traditional methods exhibit limitations in accuracy, noise resistance, and feature extraction. To address these issues, this study employs the Multi-scale Kernel Residual Convolutional Neural Network (MK-ResCNN) for EEG signal analysis. Experimental results demonstrate that the impedance of the novel electrode is approximately 12.8 kΩ at 70% relative humidity (RH), which is significantly lower than that of conventional dry electrode and close to the level of Ag/AgCl wet electrode. Moreover, the analysis of power spectral entropy and Pearson correlation coefficients (PCCs) demonstrated a strong correlation (PCCs > 0.8) between the signal characteristics acquired by the novel semi-dry electrode and the Emotiv device. It indicates that there was no significant difference in signal acquisition performance between the two electrodes. The MK-ResCNN algorithm based on the signal collected by the novel electrode has achieved a fatigue detection accuracy rate of 98.55% in the urban road environment, which is significantly better than the traditional methods. In conclusion, the combination of the novel semi-dry electrode and the MK-ResCNN methods can effectively support EEG acquisition and fatigue detection in real urban road environments.