The Study Of Driving Fatigue Analysis Method Based On Electroencephalogram Signal

In recent years,traffic safety accidents have occurred frequently,which caused widespread concern in society.Studies have shown that driving fatigue has become a direct cause of mega-traffic accidents,which not only caused significant casualties,but also brought a heavy financial burden to the victims' families.Therefore,it is very important to study the formation mechanism and physiological characteristics of driver fatigue,and then design an effective driving fatigue detection method to prevent fatigue driving and improve road traffic safety.At present,the detection method of driving fatigue is mainly based on the driver's physiological characteristics and behavioral characteristics,and the EEG signal has been regarded as the "gold standard" for detecting fatigue.The physiological characteristics based on EEG signals are considered to be effective and objective means judgment mental state.The main research contents of this thesis include the analysis of driver's EEG and EOG signals by using fusion entropy,and then classifying driving fatigue in support vector machines,and obtaining high judgment accuracy.A deep learning model based on convolutional neural network and long short-term memory network is designed.The driving fatigue EEG signal is analyzed and obtained a good classification result.The specifics are as follows:Firstly,the EEG signals of the six channels as PO3,POz,PO4,O1 h,Oz and O2 h in the occipital region of the brain and the horizontal and vertical EOG signals were selected for the analysis of this experiment.Four typical rhythmic waves(? wave,? wave,? wave,? wave)of EEG signals are obtained by discrete wavelet decomposition and reconstruction methods.Then,the approximate entropy,sample entropy and spectral entropy with each rhythm wave of EEG signals are extracted separately,and the sample entropy of horizontal and vertical EOG signals are used as features.For each entropy feature in different rhythm waves,the feature is fused in different frequency bands by the canonical correlation analysis,and finally the fatigue classification detection is performed in the support vector machine.The experimental results show that the fusion entropy feature processing is superior than the classification results of the extracted features in each rhythm wave.The fusion entropy feature classification result is 99.1%,and the method shows excellent robustness.Secondly,after pre-processing,the 24 channels EEG signal data were analyzed in the sliding window by the independent component analysis for remove the EOG artifacts interference.The abstract feature extraction is automatically performed by the convolutional neural network,and then driving fatigue classification in the long short-term memory network.The experimental data showed that the average accuracy of all subjects tested in the network model was 96.9%.It shows that the deep learning model designed in this paper is also suitable for driving fatigue detection.