Classification of motor imagery BCI using multiband tangent space mapping

Abstract

The tangent space mapping (TSM) becomes an effective method to implement brain computer interface (BCI) with motor imagery. In this paper, TSM is employed with multiband approach to extract discriminative features from electroencephalogram (EEG) to enhance classification accuracy. The EEG is decomposed into multiple subbands and the sample covariance matrices (SCMs) are then estimated on each of the subbands. Those matrices are then mapped onto the tangent space yielding the features. These obtained features of individual subbands are combined together. The dimension of the features space is reduced using principal component analysis (PCA) with one-way ANOVA. Support vector machine (SVM) based classification is performed employing the features with reduced dimension. The results of binary and four-class classification with public data sets showed that the proposed method significantly improved the performance compared to the state-of-the-art methods.