Study On Frequency Optimization And Classification Algorithm In Motor Imagery Based Brain-Computer Interfaces

Brain-computer interface (BCI) is a technology to translate human thoughts into output commands. In motor imagery (MI) based BCI, common spatial pattern (CSP) is a successful algorithm. Its advantage is devising optimal spatial filter, which can extract spatial features of electroencephalogram (EEG) signal. However, the performance of CSP algorithm depends largely on the frequency information of EEG. Hence, it is important that how to select optimal EEG signal frequency band. In order to address frequency optimization problem, in this paper, two algorithms of frequency optimization selection based CSP have been proposed.The first algorithm is wavelet packed coefficients weighted method which is appropriate for binary brain-computer interface system. Based on the frequency band theory of wavelet packet decomposition, this algorithm selects the optimal frequency band by weighting wavelet packet coefficients. The second algorithm is applied to multi-classes brain-computer interface, combined with filter bank and features selection. In this algorithm, CSP be used to extract features of every band signal; feature selection method will be utilized to select optimal frequency band information. In off-line analysis, these two proposed algorithms both yield relative better classification accuracies. Compared with broad-band algorithm, both of the above algorithms have promoted classification accuracy drastically.