Research And Implementation Of Brain-Computer Interface In Hybrid Experimental Paradigm Based On Relevance Vector Machine Classifier

Brain-computer interface(BCI),as a new type of information interaction,enables people to communicate directly with the external environment.BCI technology has broad application prospects in the fields of medical rehabilitation,education,entertainment and military.In this paper,the BCI method is studied based on noninvasive electroencephalogram(EEG).The feature extraction method based on the combination of phase space reconstruction(PSR)method and common spatial pattern(CSP)is used.A new type of composite kernel function based on relevance vector machine(RVM)classifier.A hybrid paradigm brain-computer interface(h BCI)combining motor imagery and SSVEP is designed.Online BCI system implementation is carried out.The main research contents are as follows:(1)The methods of EEG signal preprocessing and feature extraction are studied.In the preprocessing part of the EEG signal,independent component analysis(ICA)was used to extract the independent components of the electrooculogram(EOG),and the noise artifacts were removed;the PSR was applied to the analysis and processing of EEG signal,and extract CSP features in phase space(PSCSP).(2)The design of the BCI system classifier,a new composite kernel function RVM classifier is proposed.The kernel function of the RVM classifier is analyzed.A Gaussian kernel function with local characteristics and a polynomial kernel function with global characteristics are selected.On the motor imagery EEG dataset,a new composite kernel function RVM classifier is constructed by combining linear weights.(3)Research on hybrid experimental paradigm BCI.Aiming at the problem of low BCI information transfer rate(ITR)in a single paradigm,an h BCI system combining motor imagery and SSVEP was designed.Four types of motor imagery tasks(left-hand,right-hand,both feet motor imagery and none)and four types of SSVEP(6Hz,7.5Hz,8.5Hz and 10Hz)tasks are designed,and the classification of multi-task hybrid experimental paradigm BCI is studied.(4)Development and implementation of online BCI system.The hybrid paradigm experimental BCI platform was established to realize the online control experiment of NAO robot.Four types of motor imagery tasks(left-hand,right-hand,both feet motor imagery,and none)and four types of SSVEP tasks(6Hz,7.5Hz,8.5Hz and 10Hz)are designed.The motor imagery tasks control the speed of the NAO robot(in high speed v=0.5m/s or in low speed v=0.2m/s)and as a switch of the SSVEP paradigm,the SSVEP task controls the direction selection of the NAO robot.The experimental results show that the online h BCI experimental platform designed in this paper has high reliability.