Brain Computer Interface Research Based On Motor Imagery Potential

Brain Computer Interface(BCI)System based on motor imagery(MI)potential can help subjects communicate with external environment through analysising their EEG signals instead of nerves and muscles,which shows more advantages in the field of medical rehabilitation since it doesn't dependent on external stimuli.However,BCI based on MI potential possess a lower accuracy the same time requires a larger amount of leads,which significantly reduces its practicability.Researching on the reduction of leads in pattern recognition of MI,a method of wavelet packet decomposition(WPD)combined with common space pattern(CSP)is adopted to extract features more comprehensively,after which support vector machine(SVM)with parameters grid searching is applied to classify.1.Wavelet packet decomposition(WPD)and reconstruction is adopted to perform signal processing for non-stationary EEG signals with low SNR.CSP is used to extract the feature of MI from the domain of space.WPD is alpplied to extend dimensions of EEG signals for CSP algorithms thus the joint feature extraction.in the domain of time,frequency and space of MI is achieved.2.The selection of the feature dimension of CSP and lays of WPD is systematically studied.The strategy of optimizing the dimension of CSP feanture for different subjects is designed and applied in the different levels of WPD to determine the optimal number of decomposition layers.3.The classification methods of fisher linear classifier,probabilistic neural network classifier(PNN)and SVM based on kernel functions are analysed.The key parameters are optimized and the classification results are compared.Results show that the support vector machine can achieve the highest accuracy in the MI classification with less leads.4.The EEG experiment based on the MI potential was designed and conducted using the Symtop NT9200 EEG amplifier for four subjects.Appling the method of the combination of WPD,CSP and SVM,a pattern recognition effect of high accuracy in the situation of 3 leads is obtained.The validity of the proposed algorithm combination in the general environment is verified.