Multi-class Motor Imagery Classification using Multi-class SVM with Multi-band Riemannian Tangent Space Mapping

In this paper, we propose a novel multi-class motor imagery (MI) classification method in electroencephalogram (EEG)-based brain-computer interface (BCI) using multi-class support vector machine (SVM). EEG signal is decomposed into multi-band signal and then for each sub-band, spatial sample covariance matrix is computed. By applying Riemannian tangent space mapping method which utilizes the geometric structure of covariance matrices to the sub-band spatial covariance matrices, sub-band features are extracted and combined to form a feature vector. In order to improve multi-class classification performance, the feature vector is passed into multi-class SVM which directly tackles multi-class problem, in contrast to the existing works where one-versus-one or one-versus-rest strategy is used to indirectly solve multi-class classification problem. The performance of the proposed method is evaluated on the 4-class BCI Competition IV dataset 2a and the experimental results show that the proposed method improves the mean classification accuracy.