Classification of Motor Imagery EEG Signals using MEMD, CSP, Entropy and Walsh Hadamard Transform

Abstract

This paper provides a novel set of features for classification of motor imagery tasks including the following two classes: right and left hand. While performing motor imagery tasks, desynchronization is seen in the mu and betabands over the sensorimotor cortex region. In order to capture these changes in the different frequency bands, we use MEMD for decomposing the EEG into oscillatory components called IMFs which characterize either a single frequency or a narrow band of frequencies. Features are extracted by applying common spatial pattern (CSP), Entropy and Fast Walsh Hadamard Transform (FWHT) on these IMFs. Using SVM classifier, the above features yield a maximum accuracy of 95%. The proposed feature set results in a better discrimination for motor imagery signals compared to the earlier work in this field.