Optimal channel selection using covariance matrix and cross-combining region in EEG-based BCI

The EEG-based brain-computer interface (BCI) requires removal of irrelevant channels to improve performance. In this paper, we propose the optimal channel selection using EEG channel covariance matrix and cross-combining region. First, the discriminative H channels and target channel are selected by difference of EEG channel covariance matrix between two classes. Second, we configure several sub-channel regions to cover the H channels. Then, we extract FBCSP features from cross-combining regions which are combination of the sub-channel regions and target channel. We select the best one cross-combining region and the optimal channels which are included in selected cross-combining region are finally selected. The features of selected region are used as input of LS-SVM classifier. The simulation results show the performance improvement of proposed method for BCI competition III dataset IVa by comparing the conventional channel selection methods.