EEG Classification for MI-BCI using CSP with Averaging Covariance Matrices: An Experimental Study

Abstract

To assist disabled people by controlling an external system by using motor imagery (MI) is a common applications of brain computer interface (BCI) field. This paper we focused on an experimental comparison of covariance matrix averaging ways of EEG signal and EEG classification of two types of MI tasks $(right-hand^{\ast}$ foot and right-hand*left hand). Indeed averaging covariance matrices of EEG signal might be a used in brain computer interfaces (BCI) with common spatial pattern (CSP) method. Structured into trials is a usually paradigms of BCI which we have a tendency to use this structure into account. In addition, covariance matrices with non-Euclidean structure should be consideration likewise. We review much method for averaging covariance matrices in SVM from literature and observe through the experimented result using publicly available four datasets. Our experimental result show that for the case of averaging covariance matrices using Riemannian geometry with small dimension feature issue improve the classification performance. Our result shows the performance increase (2% >performance), but also the limit of this method once the increase feature dimension.