Analysis Of Vigilance Based On EEG Signal In Simulated Driving Environment

Vigilance is usually defined as the sensitivity or ability of people to maintain attention and alertness to external stimuli,when they perform some tasks during a long time.Vigilance is a physiological indicator closely related to the waking or fatigue state.People’s reaction time to external changes or responding accuracy for a long period can be used in evaluation of the degree of people’s alertness levels.When people feel fatigue or sleepiness,vigilance decreased significantly.This is very dangerous for vehicle drivers or pilots,and is likely to cause accidents.Now more and more people are devoted to the study of fatigue driving,and actively looking for ways to solve it,which is an important subject in the field of vehicle and aircraft-assisted safe driving,and has important practical significance to solve the problem of safe driving.The analysis based on the physiological signal is a common method to study the vigilance,usually using the physiological signals such as EEG,EEG,ECG,especially EEG,which can objectively reflect the characteristics of electrical activity of the brain and make it an important means of fatigue detection.With the development of sensors and brain-computer interfaces,EEG acquisition becomes more and more convenient.In this thesis,vigilance is divided into two states:awake state and fatigue state,and the focus is to analyze the vigilance of the driver based on the EEG signal.The main contents of this study include the design of simulated driving experiment,the acquisition and de-noising of EEG signal,the feature extraction algorithm,the choice of key frequency bands and channels.The main work of this thesis is as follows:(1)Design of simulated driving experiment and acquisition of EEG signal and its artifact removal.This thesis uses the Emotiv Epoc system to collect and record 14 channels of EEG data in a simulated driving environment.The filter,independent component analysis and wavelet de-noising method are used to remove artifacts such as power frequency power,electromyogram and electric signal.(2)Extracting the traditional features of EEG signals,using wavelet transform method to obtain four kinds of rhythmic waves,namely,delta,theta,alpha,beta wave.Then calculating the energy and entropy of each frequency band signal.Secondly,this thesis mainly uses the deep belief network and convolution neural network to automatically learn the features of EEG,and tries to extract the abstract features which are difficult to be found by artificial calculation method.This method overcomes the limitation of manual extraction.(3)In order to remove the EEG signals with little correlation with vigilance,the key rhythms and key channels are selected from the 14 channels,and are classified and identified.And then compared with the classification results of whole channel and full rhythm EEG signals.(4)Support vector machine is used as the classifier to classify the EEG signals by using the traditional features and the ones learned by deep networks as input.At the same time,the deep networks are also used for classification.By comparing the results of classification,it is found that the feature learning and classification based on deep belief network is the best,and the accuracy rate is up to 94.97%.Compared with the classification accuracy based on traditional features,it has been greatly improved.(5)Finally,based on the classification of EEG signals,the feasibility of using EEG to analyze drivers’ vigilance is verified.