Research On EEG Feature Selection And Feature Extraction Algorithm Based On Motor Imagery

The research of Brain-Computer Interface(BCI) technology is widely highlighted and developed rapidly over the last decade, which it provides a new way of information exchange and control access to the brain and external environment. As a main branch of BCI, the motor imagery EEG is simple in operation and easy to design, but also has the problem of classification accuracy and individual differences. So this thesis selects the motor imagery as the research object and do research on the classification algorithm for EEG.For more channel and frequency movement imagine EEG, its feature extraction algorithm using largely depends on the feature selection. So the selection of main parameters of the original EEG before the feature extraction is necessary. We proposed the feature selection algorithm based on the correlation coefficient and divergence analysis are proposed after analyzing the traditional one of the mutual information.Use the third international BCI competition data sets of small sample learning motor imagine do the feature extraction of the AR model spectrum estimation and common spatial pattern(CSP) algorithms after three algorithms feature selection, and classification by the LDA classifier. The simulation results show that the correlation coefficient and divergence analysis algorithms can be good choices for optimal frequency and channel parameters,and have superior ACC and MI values than the mutual information algorithm.Finally, it combines feature of the AR and CSP algorithms to classify, the resulted show that the CSP-AR feature extraction algorithm is more practical.Research shows that using the divergence analysis of feature selection algorithm and do the CSP and AR feature extraction combines, finally get the classified output through the LDA classifier can obtain the maximum averages of ACC and MI. So this experimental scheme is more suitable for the classification of motor imagery EEG.