| Brain-computer interface(BCI)is a direct communication and control system between the human brain and external device without using traditional pathway consisting of muscles and peripheral nerves,and it is also a human-computer interaction way based on electroencephalogram(EEG).In the BCI system,the Motor imagery EEG signal,which is generated by different limb movement,can be converted to the corresponding control commands to control the external devices through analysis and processing.According to the characteristics of the Motor imagery EEG signal,the process of the de-noising,feature extraction and optimization,feature classification has been researched and analyzed on the two types of motor imagery EEG signal.In the paper,the main content and the innovation are as follows:⑴ In order to eliminate the noise mixed in EEG,the paper puts forward a new EEG de-noising method based on ensemble empirical mode decomposition(EEMD)and improved wavelet threshold method.New threshold function and threshold selection rules are introduced in the improved wavelet threshold based on the traditional wavelet threshold.Firstly,the EEG is decomposed by the EEMD.Then,according to the noise after decomposing is mainly distributed in the first few high-frequency,so the few high-frequency components has been chosen for the improved wavelet threshold method analysis.Finally,the processed high frequency components and low frequency components are reconstructed to get the signal after de-noising.The de-noising algorithm is a good way to eliminate noise in EEG and keep most of the useful information,laid a good foundation for subsequent processing.⑵ Motor imagery EEG contains time,frequency and spatial domain information.A single feature extraction methods cannot extract the features which can comprehensively reflect the characteristics of the signal.In this paper,empirical mode decomposition(EMD)and wavelet packets that can extract frequency-domain information respectively combined with common spatial patterns(CSP)that can extract spatial-domain information for extracting EEG features.The two combination method both can extract the multi-domain feature of EEG signal well,and can be helpful for the subsequent classification.⑶ Different individuals has different motor imagery EEG.For multi-channel EEG data,a new channel selection method is proposed which compute average energy difference about two types of imagine tasks of different channels.This method guarantee the maximum utilization of the data and reduce the number of channels,so it improve the efficiency of the process.In addition,considering the extracted features may have less effect on classification and even bring some interference,a feature selection method is proposed based on decision tree.Through constructing decision tree,the most important characteristics for classification can be find out.Correspondingly,the accuracy and efficiency of classification is greatly improve.⑷ Support vector machine(SVM)which is established model based on radial basis kernel function is used to classify the EEG signals.In order to improve the SVM performance,the cross validation(CV)algorithm and genetic algorithm(GA)are used to optimize the penalty factor and kernel parameters.Experimental results show that these two methods can get good classification results.In addition,the classification of the former is better,and the latter is more efficient.⑸ Choose the simulated EEG data,the 2003 BCI competition Datasets3 data and the 2008 BCI competition Datasets1 data as simulation experiment data.Experiment results to prove that this paper studies the algorithm effectively,and to a certain extent,also compared the advantages and disadvantages of different feature extraction and classification algorithm. |
PDF Full Text Request