Electrocorticogram classification based on wavelet variance and Fisher linear discriminant analysis

For a typical electrocorticogram(ECoG)-based brain-computer interface(BCI) system, a pattern recognition algorithm using wavelet analysis and Fisher linear discriminant analysis(FLDA) was proposed. Firstly, based on studying wavelet theory, a novel feature extraction method in ECoG signal processing namely wavelet variance(WV) or wavelet packet variance(WPV) was proposed considering the band interlacing phenomenon in wavelet packet transform, and the computing method of WV/WPV was brought out; then, taken as feature, the WVs and WPVs of 6 most important channels were selected from 64 channels for analysis, consequently the ECoG data were three-layer decomposed, the WVs and WPVs containing Mu rhythm and Beta rhythm were taken out as final features based on ERD/ERS phenomenon; finally the final features were classified with FLDA in optimum-intervals of the ECoG data. The results showed that the max accuracy for test data was 92%, wavelet variance and wavelet packet variance could be taken as efficient features for ECoG.