Toward Practical Driver Fatigue Detection Based on EEG Using Forehead Low Channel and Cascaded Deep Forest

Objective Performance decrement associated with fatigue is a leading contributor to traffic accidents and fatalities. Electroencephalogram (EEG) is generally accepted as one of the reliable physiological indicators but wearing a multichannel headset to acquire signals limits the application of EEG-based interactive device among drivers. Prefrontal EEG is a potential candidate channel for early detection due to the rise of portable wearables. Methods In this work, an applicable, efficient and robust driver fatigue detection method for practical application was proposed by using forehead EEG based on a deep forest (DF) framework. Incorporating two rapid and effective features of wavelet log-energy entropy and high-order component statistics was beneficial for uncovering valuable hidden information within single-channel EEG. A comprehensive labeling engineering of the drivers' fatigue state was conducted across 26 subjects. Results The experimental results showed the better expected outcome was obtained than that of previous studies by hitting an average accuracy of 95.1%. In addition, the energy information of the small-scale oscillations of brain signals from different frequency bands and the application value of higher order statistics of the nonlinear dynamic state from the reconstructed phase space was further verified in real-time signal processing, especially in low-channel signal feature extraction. Significance This study presented a new framework of using prefrontal EEG based on a cascade structure to construct a landing fatigue detection method. It could also provide a potentially valuable way for biomedical signal processing in low-channel systems.