A Novel Feature Extraction Framework for Four Class Motor Imagery Classification using Log Determinant Regularized Riemannian Manifold

Brain-Computer Interface (BCI) systems allow the person in communicating with the external world using Electroencephalography (EEG). Motor Imagery (MI) based BCI systems play a vital role in interacting with the external environment. In this paper, we propose a novel robust feature extraction and classification framework for four class MI classification to improve the classification accuracy. The proposed architecture is developed using log-determinant (log-det) based Regularized Riemannian mean (LDRRM) and linear SVM. The robustness of features extracted from the four class MI data is improved to the outliers and noise by using the proposed LDRRM framework. We evaluated the performance of the proposed LDRRM classification framework on publicly available four class MI dataset 2a of BCI competition IV. The performance results show that the proposed LDRRM classification architecture obtained a mean classification accuracy of 69.12%, also achieved 1.54% higher classification accuracy when compared with the existing studies.