Research On Train Driver Fatigue Detection Based On EEG Signal

Fatigue driving is one of the main causes of train accidents,causing significant casualties and economic losses.It is an urgent problem to timely and accurately detect the fatigue state of train drivers and assist them in reducing fatigue driving behavior.To this end,research on fatigue detection algorithms based on electroencephalogram(EEG)signals has been carried out,with the following main research contents:(1)A fatigue state induction and EEG signal acquisition experiment was designed for data collection.Based on this,EEG signals in both awake and fatigue states of the subjects were collected,forming a dataset related to fatigue driving.The EEG signals were preprocessed,mainly including filtering and artifact removal.(2)Power spectral density features and differential entropy features were extracted from the frequency domain and time domain,respectively.Based on the extracted features,traditional machine learning classification methods were studied.Three machine learning models,including LDA,KNN,and SVM,were built and compared in both within-subject and mixed-data experiments.In the within-subject experiment,the SVM model achieved an average classification accuracy of 0.83,which was 0.09 and 0.04 higher than LDA and KNN,respectively.The average AUC value was 0.89,which was 0.07 and 0.04 higher than LDA and KNN,respectively.The average F1 score was 0.82,which was 0.10 and0.04 higher than LDA and KNN,respectively.In the mixed-data experiment,the SVM model had better comprehensive performance in the five evaluation criteria of classification accuracy,AUC,F1 score,precision,and recall.(3)A CNN-LSTM-SVM EEG fatigue detection classification model was proposed based on deep learning and combined with traditional machine learning methods.With the deep learning method,the process of manually selecting and extracting features could be skipped,and the deep features of EEG data could be learned,including some features that could not be learned manually.In the within-subject experiment,the CNN-LSTMSVM model achieved an average accuracy of 0.94,an average AUC value of 0.96,and an average F1 score of 0.94,which were 0.05,0.01,and 0.05 higher than the CNN-LSTM,CNN,and SVM models,respectively.In the mixed-data experiment,the CNN-LSTMSVM model had better performance in the five evaluation criteria than the other three models,with accuracy and AUC reaching 0.93 and 0.98,respectively.(4)A Transformer fatigue detection regression model was proposed to fit the eyelid closure degree curve based on EEG signals,so as to realize the judgment of fatigue state and the visualization of its dynamic changes.In the regression experiment,the Transformer model achieved an average EVS,R2,RMSE,and MAE of 0.92,0.90,0.06,and 0.05,respectively,which were better than those of the SVR,RNN,and LSTM regression models used as comparisons.