SVM based Classification Of sEMG Signals using Time Domain Features for the Applications towards Arm Exoskeletons

Abstract

An exoskeleton robot performance can be improved by providing accurate control commands using information from the surface EMG signals. This paper proposes a classification of the hand movements based on sEMG signal. We explore the time domain and time frequency domain features from which a set of selected features are provided to a multi-class SVM classifier. Finally, the proposed method is evaluated on a benchmark-scientific database, the NINAPro-DB1, consisting of 52 sEMG hand movement classes obtained from 27 subjects. The average classification accuracy of 84.4% has been achieved for 52 classes using 10-fold cross validation method.