Research On The Construction Method Of Functional Brain Network Based On EEG State Recognition Of Fatigue Driving

Functional brain network(FBN)is an intuitive expression of dynamic neural activity interaction between different neurons,neuron clusters or cerebral cortex regions.It can characterize the topological structure and dynamic characteristics of the brain network.It is a challenging subject to choose a suitable method to construct FBN to accurately and effectively characterize the characteristics of the brain network.For the electroencephalograph(EEG)data set,the research on the construction of functional brain networks that meets the needs to obtain the corresponding brain network types and characteristics plays an important role in the detection and recognition of specific needs of the brain state.The subject aims to improve the recognition accuracy and stability of the FBN-based fatigue driving state recognition model by exploring two FBN construction methods.The mainly research content and the phased results obtained are as follows:(1)Research on FBN construction method based on entropy.Entropy can effectively describe the complexity,nonlinearity and uncertainty of EEG signals.It is of great significance to study the FBN structure based on entropy.Firstly,the subject studied in detail the use of four entropy features to characterize EEG signals,including fuzzy entropy(FE),sample entropy(SE),approximate entropy(AE)and spectral entropy(Sp E).Secondly,this subject studies the method of constructing the corresponding FBN based on these entropy features,constructs the FBN based on fuzzy entropy,sample entropy,approximate entropy and spectrum,as well as constructing the fatigue driving state recognition model based on the corresponding FBN(abbreviated as separately FE＿FBN,SE＿FBN,AE＿FBN,Sp E＿FBN).Finally,through the analysis of network measurements,experiments on the data set show that good recognition accuracy and recognition stability can be achieved based on the FE＿FBN model.Compared with other models based on the same data set,our model can obtain higher accuracy and more stable classification results even if the length of the intercepted EEG signals is different.(2)Research on the FBN fatigue driving state recognition model(abbreviated as KCCA＿FBN)based on kernel canonical correlation analysis(KCCA).At the present stage,the research on the recognition of EEG noise basically uses the cumbersome and time-consuming method of noise reduction,denoising and then signal reconstruction.This subject is dedicated to studying the ability to adapt to EEG noise data without reconstructing the signal.The KCCA-based measurement method is proposed to describe the strength of the connection between different network nodes,and the KCCA＿FBN fatigue driving state recognition model is constructed.The purpose is to realize that the model constructed by training pure EEG data can be used for noisecarrying EEG.The accuracy of data detection has the effect of significantly improving.This method is used to analyze the correlation between two sets of nonlinear EEG data to find a set of representative comprehensive variable pairs,and make the correlation coefficient of the variable pairs mapped in one-dimensional space the largest.This largest correlation coefficient not only represents the correlation of these two sets of multidimensional variables,but also represents the weight of the most representative characteristics of these two sets of variables in all typical variable pairs.Experiments have confirmed that the functional brain network model constructed by this method with pure EEG signals and noise-carrying EEG signals have similar local efficiencies.By testing on the fatigue driving data set,using the KCCA＿FBN model trained on pure EEG data can classify and recognize the EEG data carrying noise,and has achieved high recognition accuracy and strong generalization ability.