Research On Channel Selection Method And Its Application Of The Motor Imagery EEG

The brain-computer interface(BCI)system establishes a direct connection between the brain and external devices,and can directly decode brain activities into control instructions for external devices by recognizing brain signals.The BCI system is increasingly used in military,education,medical,entertainment and so on.The BCI system based on motor imagery is more consistent with people's natural imagination and motor control,so motor imagery electroencephalography(EEG)is researched in this paper.Due to the small sample size and high signal dimension of motor imagery EEG,channel selection appears to be particularly important.Excessive electrode channels,on the one hand,will increase the cost of the equipment,and on the other hand,will reduce the comfort of users,which is not conducive to the promotion and use of the equipment.This paper mainly studies the method of channel selection and its application.The specific research contents are as follows:1.The existing work mainly studies the improvement of channel selection methods,neglecting the applicability of different features to different channel selection methods.Therefore,the influence of feature extraction on channel selection methods is systematically studied.The extracted features are variance,autoregressive coefficient,band-pass power,wavelet packet energy and the fusion features of the four features.The compared channel selection methods are Fisher discriminant criterion(FDC),recursive channel elimination based on support vector machines,least absolute shrinkage and selection operator(LASSO)and Group LASSO.The experimental results show that no matter which channel selection method is used,the classification effect of fusion feature is better than that of single feature.When extracting fusion features,the average accuracy of all subjects and methods was77.03%.2.Aiming at the noise sensitivity and overfitting problems of the traditional common spatial pattern(CSP),an improved CSP method based on channel selection is proposed.On the one hand,channel selection can eliminate channels containing noise and redundant information and reduce CSP noise sensitivity.On the other hand,channel selection can reduce the estimation bias of sample covariance,thus relieving the overfitting problem.In this method,the variance of each channel signal in EEG is extracted as the feature,and the channel weights are obtained by Fischer linear discriminant analysis and Bayesian linear discriminant analysis,respectively.After sorting the channel weights,the wrapper method is used to select the optimal channel combination.Experimental results show that channel selection can significantly improve CSP classification performance,the result of BLDA classification reached 87.77%.3.The method which based on wavelet packet decomposition(WPD)and CSP(WPDCSP)make up for the frequency defect of CSP method,but it is very time consuming.To solve the time consuming problem of WPD-CSP,an improved WPD-CSP method based on channel selection is proposed.Firstly,the band-pass power features are extracted.Then the subject-specific optimal channel is selected by FDC.Next,the selected channels are decomposed by WPD,and the frequency subbands related to the motor imagery task were selected for CSP feature extraction.Finally,Fisher linear discriminant analysis is used for classification.Experimental results show that the proposed method has good classification accuracy and reduces the time of feature extraction.The accuracy of 83.11% is achieved on the used public data sets,and the time of feature extraction is reduced to 1/7 of the original time.The research content of this paper,on the one hand,can improve the classification performance of motor imagination EEG,and on the other hand,can improve the comfort and portability of BCI system.