Reaserch Of Feature Extraction And Classification For Motor Imagery EEG

Brain – Computer Interface(BCI)is a new technology which has attracted more and more attention of the community with the development of science and technology.In BCI system,the electrical activity of the cerebral cortex can be detected through the electrode cap and other equipment.Then,the motor imagery EEG signal can be converted to instructions to control the external device so as to communicate with the outside world.So the EEG signal can be used to control the external auxiliary equipment,which can bring the opportunity for the people with certain disabilities to re-communicate with the external environment.In the paper,the research is focused on the two class of motor imagery EEG signal processing.As the poor discriminant of the extracted feature and low classification rate for EEG,and according to charactericstic of nonlinear and rhythm of EEG signal,some methods have been presented in elimiting the noise in EEG and improving the discriminant of the feature,and optimizing the classifier,to improve the classification accuracy of the motor imagery EEG signal.The main research work of the paper has been arranged as follows:(1)The EEG signal preprocessing: an improved soft threshold method has been adopted to elimate the noise signal on the basis of soft threshold.In comparision with the traditional hard threshold and soft threshold with the index of SNR and MSE,the results show that the improved soft threshold method not only can retain the advantage of the original method,but also overcome the shortcomings to improve the signal-to-noise ratio.(2)The feature extraction of motor imagery EEG: since the nonlinear and non-stationary of the EEG signal and due to the ERD/ERS phenomena during motor imgery,a method combining wavelet transform with fuzzy entropy has been proposed for motor imagery EEG feature extraction.The EEG signal of the channel of C3 and C4 has been decomposed using wavelet transform.Then,the alpha rhythm and beta rhythm signal can be extracted by the algorithm of fuzzy entropy.At last the data from 2008 BCI competition has been used to analyze the validity and feasibity.(3)The pattern classification of motor imagery EEG: in order to solve the problem of kernel parameter selection in support vector machine(SVM),particle swarm optimization(PSO)algorithm has been adopted to optimize the penalty parameter C and the kernel parameter g to improve the performance of the classifier.Then,the optimized classifier has been used for classification and prediction of motor imagery EEG signal.Also the classifation theory and the basic processes of the particle swarm optimization optimizing the SVM classifier has been presented in detail.(4)The experimental results and anylysis: in this part,the competiton data from BCI Competition 2005 data IVa and 2008 BCI Competition IV Datasets1 have been used to conduct experiment about motor imagery EEG feature extraction and pattern recognition.In experiment one,the final resuls show that the PSO optimized SVM classfier can improve the classification accuracy of EEG signal effectively.In experiment two,the wavelet transform combined with fuzzy entropy method has been used for EEG feature extraction,and the PSO optimized SVM classifer has been adopted for pattern recognition.The final results demonstrate that the wavelet fuzzy entropy algorithm can effectively extract the EEG signal feature,and the classifaction rate of PSO optimized SVM classifier is higher than the traditional SVM.