Research On Multi Pattern Recognition Method And Application Of Motor Imagery Brain-Computer Interface

Motor imagery(MI)brain computer interface(BCI)is a technology based on the active imagination of EEG,and realizes the communication and control channel between human brain and computer or other electronic devices by using external auxiliary devices.In this paper,the signal processing algorithms in BCI system,includes filtering algorithm,feature extraction algorithm and feature classification algorithm,are studied.A new kernel function relevance vector machine((RVM))classifier is proposed.Finally completed the four-rotor control experiment of online BCI system.The main contents of this paper are as follows:(1)Three kinds of EEG signal filtering methods are analyzed,including Butterworth filter,common average reference filter and Laplacian filter.The Butterworth filtering method of 3-24 Hz is selected to remove the high frequency interference and noise artifacts through the analysis of the frequency feature of the motor imagery EEG signal.The one versus one common spatial pattern(OVO-CSP)is designed to extract the features of four-class motor imagery EEG.(2)The advantages and disadvantages of the classic EEG feature classification algorithms,linear classifier,support vector machine(SVM)and artificial neural network(ANN),are analyzed.To avoid the shortcomings of SVM,relevance vector machine based on Bayesian sparse probability model is selected as the classification algorithm for motor imagery EEG.(3)The kernel function of relevance vector machine is improved,and a new kernel function relevance vector machine classifier is proposed.Firstly,the conventional relevance vector machine classifier is improved by using the combined kernel function instead of the single kernel function.The experimental results show that the combined kernel-based RVM classifier has higher classification accuracy than the single kernel-based RVM classifier.Then,a new chaos kernel function relevance vector machine is proposed,aiming at the dynamic characteristics of EEG signals.The experimental results show that the proposed classifier has more accurate EEG classification accuracy compared with the Gaussian kernel function RVM classifier.(4)In order to verify the effectiveness of the designed method,an online BCI platform is built.Achieve to control the four-rotor four movements,taking off,forward,left shift and landing,by four-class motor imagery tasks effectively.The online experiment confirms that the designed online BCI system has high reliability.