Motor imagery classification using subband tangent space mapping

This paper presents a novel approach of motor imagery (MI) classification using subband implementation of tangent space mapping (TSM). The multichannel electroencephalography (EEG) signals are decomposed into five subbands. The sample covariance matrix (SCM) of individual subband is projected to tangent space using TSM yielding the tangent features. Thus obtained features space has a high dimension. The principle component analysis (PCA) is employed to reduce the dimension of the feature space based on the p-value of one-way ANOVA. The classification is performed by support vector machine (SVM) with reduced dimension. The experimental results with publicly available datasets show that the proposed method significantly improves the performance motor imagery classification.