Eeg-based Mental Fatigue Detection Based On Portable Acquisition Equipment

Fatigue is a kind of subjective discomfort always caused by longtime,high-intensity physical labor or mental work. With the ability of concentrating, responding and judging quickly decreasing, fatigue has a pernicious influence on our daily life and safety. In some severe cases,fatigue may even lead to chronic fatigue syndrome and other chronic diseases. Therefore it is essential to detect fatigue in time and prevent its harm.Electroencephalogram (EEG) is a kind of spontaneous biological electrical signal generated by human body, which could directly reflect the activities of human brain neurons. And EEG has been proven to be a robust indicator of human cognitive states.This paper focuses on EEG-based mental fatigue detection system. The main contributions of this paper includes four parts.1. Most of the existing researches related to EEG applications are based on multi-channel EEG signals. And the acquisition equipment mostly applies wet electrodes, which are relatively expensive and might need complicated operations. This experiment uses a piece of portable acquisition equipment with dry electrodes to construct an EEG dataset on a scale of 100 volunteers.There are three kinds of mental states included in the dataset, alert, slight fatigue and severe fatigue. The database is helpful in model training,testing and the evaluation of mental fatigue detection algorithms.2. This paper designs two adaptive methods for raw EEG signal preprocessing.One method applies a 3Hz Butterworth high pass filter and a 50Hz notch filter which could meet real-time and low computational complexity requirements.The other method applies EEMD method to remove the components of noise based on the first method, which could meet the requirement of higher accuracy.3. This paper analyzes EEG signals and the corresponding mental states, builds a frame mixing time domain analysis, frequency domain analysis, frequency domain analysis and nonlinear dynamic analysis methods on the preprocessed EEG signals, extracts 54-dimensional features and completes feature selection for the subsequent classification.4. Using the extracted features, this paper proposes a mental fatigue detection algorithm based on DBN (Deep Belief Network). The proposed DBN has the structure of four layers of 20-50-10-3 nodes and achieves the accuracy rate of 92.55% on test dataset. Compared with some linear or non-linear traditional classifiers and some boosting methods, the experimental results show that the proposed DBN-based model has higher accuracy and better general applicability.