Abstract:Establishing a mental fatigue monitoring system is of great importance as for severe fatigue may cause unimaginable consequences. Electroencephalogram (EEG) is often utilized for mental fatigue detection because of its high temporal resolution and ease of use. However, many EEG-based approaches for detecting mental fatigue only take into account the feature extraction of a single domain and do not fully exploit the information that EEG may offer. In our work, we propose a new algorithm for mental fatigue detection based on multi-domain feature extraction and fusion. EEG components representing fatigue are closely related in the past and present because fatigue is a dynamic and gradual process. Accordingly, the idea of linear prediction is used to fit the current value with a set of sample values in the past to calculate the linear prediction cepstral coefficients (LPCCs) as the time domain feature. Moreover, in order to better capture fatigue-related spatial domain information, the spatial covariance matrix of the original EEG signal is projected into the Riemannian tangent space using the Riemannian geometric method. Then multi-domain features are fused to obtain comprehensive spatio-temporal information. Experimental results prove the suggested algorithm outperforms existing state-of-the-art methods, achieving an average accuracy of 87.10% classification on the public dataset SEED-VIG (three categories) and 97.40% classification accuracy (two categories) on the dataset made by self-designed experiments. These findings show that our proposed strategy perform more effectively for mental fatigue detection based on EEG.

EEG-based mental fatigue detection using linear prediction cepstral coefficients and Riemann spatial covariance matrix