Research Of Critical Channels And Rhythms In EEG Based Driving Fatigue Detection

Traffic accidents caused by driving fatigue account for a large proportion in all traffic accidents,so it is very important to find an objective and accurate fatigue detection method to avoid its impact on safe driving.Among various detection methods,electroencephalogram(EEG)is known as the "gold standard",because it can directly reflect the neural activity of the brain.However,due to the specificity of EEG,it is still difficult to estimate the fatigue state quickly and accurately.In addition,related studies lack the analysis of the critical EEG channels and rhythms for fatigue detection,while the redundant signal channels have large pressure and interference for the signal acquisition and processing system,which is not conducive to its application in the engineering.Focusing on the above reasons,this thesis intends to study driving fatigue detection based on EEG signals,and analyze the critical EEG channels and rhythms related to driving fatigue.The specific research contents are as follows:Based on the results of a large number of literature research,there is no publicly available multichannel EEG dataset collected from the whole brain that is suitable for this study.Therefore,this thesis systematically summarized the popularly used driving mental fatigue inducing experiment protypes,then designed a fatigue driving inducing experiment that can be carried out in the laboratory environment,and constructed the EEG dataset for fatigue detection in this thesis.Considering that the neuron activity and EEG signal of the brain are nonlinear dynamics and low-dimensional chaotic characteristics,the more accurate classification method of EEG signal should be to extract its nonlinear features,a fatigue detection method based on multiple nonlinear features fusion is proposed.Through multiple kernel learning(MKL),six nonlinear features are optimized and fused.Experimental results show that the recognition accuracy of this method is higher than that of the traditional feature direct connection method.After that,a key EEG channel analysis method based on the optimal nonlinear feature and deep belief network is presented.Since traditional EEG features only consider the amplitude information of EEG,which can not describe the whole brain function,a fatigue detection method based on the theory of brain network is proposed.Firstly,based on the multi-channel EEG signal,the partial directed coherence(PDC)is used to measure the functional connection characteristics between different electrodes,and the directed binary brain functional network is established.After that,the network related statistical features are extracted,and the fatigue detection model is established.Finally,the critical EEG channel analysis method based on the brain network feature of betweenness centrality is proposed.Considering the tedious work of traditional feature extraction based on expert experience and taking the full use of convolutional neural network for classification task,this thesis proposes the concept of electrode frequency distribution maps(EFDMs)based on EEG signal,and designs a deep convolution neural network including four residual blocks for fatigue detection.Then,based on the attention mechanism of convolutional neural network,the electrode channels and frequency bands that are most related to alert and fatigue state recognition are automatically obtained from the EFDMs by gradient-weighted class activation mapping(Grad-CAM).In view of the complex learning process of deep network,which often needs a large number of labeled data for model training,this thesis also studies the driving fatigue detection based on deep model transfer,and realizes the fatigue detection with a small amount of training samples and deep pre-training model.In order to realize driving fatigue detection based on EEG,this thesis designed a simulated driving experiment scheme,constructed an EEG fatigue detection dataset.According to the characteristics of EEG signal,this thesis carried out the fatigue detection research from three aspects of nonlinear features,brain functional network and convolutional neural network,and obtained the critical EEG channels and frequency bands under the corresponding type of features.The work of this thesis will help to simplify the fatigue detection system and promote the driving fatigue detection system based on EEG signal to the practical application field.Based on the obtained key EEG channels and frequency bands,it can also fundamentally help to understand the processing mechanism of fatigue state detection system and find the neural characteristics related to fatigue state in the brain.In addition,although the goal of this thesis is to detect driving fatigue,the relevant experimental design methods and algorithms involved are also applicable to other kinds of mental fatigue detection.