EEG Signal Classification Method Based On Support Vector Machine/support Matrix Machin

As one of the main causes of major traffic accidents,fatigue driving brings great harm to society,families and individuals.Electroencephalography(EEG)records the electrical activity of nerve cells in the human cerebral cortex,which can directly reflect the immediate state of the brain and avoid the influence of human subjective factors.So it is considered to be the most effective method for detecting mental status.Therefore,it is of great significance to reduce the loss of life and property due to fatigue driving by classifying the EEG signal and thus detecting the driver’s fatigue state for further accurate early warning.In this paper,we study the EEG fatigue signal classification methods based on Support Vector Machine(SVM),Least Squares Support Vector Machine(LSSVM),Support Matrix Machine(SMM),and Sparse Support Matrix Machine(SSMM),and the main work of this paper includes the following aspects:1.To address the problem of classification accuracy of SVM algorithm,an improved SVM algorithm based on Genetic Algorithm(GA)is proposed to classify the fatigue signal of EEG.The GA algorithm is used to find the hyper-parameters in the SVM algorithm,so as to solve the problem of low accuracy of the SVM algorithm and the influence of hyper-parameters,and to improve the classification accuracy of the algorithm to a certain extent.2.A least-squares support vector machine algorithm based on improved kernel function and Particle Swarm Optimization(PSO)is proposed to address the computational accuracy and speed problems of LSSVM algorithm.Firstly,the PSO algorithm is used to optimize the parameters of the LSSVM algorithm in order to improve the classification accuracy of the original algorithm;then the improved kernel function is obtained by expanding the Gaussian kernel function of the original algorithm in Taylor series and taking the first three terms,which improves the computational speed of the algorithm.3.To address the location-dependent and time-dependent problems of EEG signals,an Auto-Correlation Function(ACF)-based SSMM algorithm is proposed to optimize and classify EEG fatigue signals.In this method,the redundant features are compressed by inputting matrix EEG signals with location information using the sparse principle;the original input matrix is extended by transforming and adding data through the ACF method,which is used to reflect the information containing the influence of previous moments on the current moment to accurately represent the memory dynamics of EEG signals.