Automated Mental Fatigue Detection From Electroencephalogram Using Joint Time-Frequency Representation Based on Multivariate Iterative Filtering

Driver drowsiness detection is a crucial technology for enhancing road safety and preventing accidents caused by fatigue. This article proposes a method for analyzing electroencephalogram (EEG) signals during driving tasks to assess the driver's mental state. Due to the nonstationary nature of EEG, the conventional Fourier spectrum is not well suited for spectral estimation of EEG. To address this, the study employs a multivariate iterative filtering (MIF) technique to decompose multichannel EEG signals into narrowband amplitude-frequency modulated components. The instantaneous amplitude and frequency are estimated using the discrete energy separation algorithm (DESA), and a joint time-frequency representation (JTFR) based on DESA is applied to estimate the spectral content of multichannel EEG. Mental states associated with drowsiness are identified using the joint marginal spectrum and an artificial neural network classifier. The proposed MIF-based framework was validated on two EEG datasets, achieving classification accuracies of 95.03$\pm$1.08% and 98.33$\pm$1.51%, respectively. These results demonstrate the potential of the method in preventing accidents caused by drowsy or distracted driving.