Pattern recognition of sEMG signals using DWT based feature and SVM Classifier

Abstract

Surface electromyogram (sEMG) signals have been widely utilized for various robotics and
medical engineering applications. In robotic rehabilitation, sEMG signal is used for designing
the myoelectric prosthetic devices. This study investigated the usefulness of discrete wavelet
transform (DWT) based feature extraction from multiple levels of approximation and detailed
coefficients obtained from sEMG signals for controlling the robotic arm prototype. DWT is
employed for de-noising as well as feature extraction process. Daubechies 2 wavelet (db2) was
employed at third level decomposition of sEMG signals. Feature vector was formed by extracting
the useful features from the third level approximation and detailed coefficients and further tested
by using Support Vector Machine (SVM) classifier. The performance of SVM classifier is also
compared with Artificial Neural Network (ANN) classifier. The classification accuracy of SVM
classifiers was found better as compared to ANN in term of speed and robustness which show the
usefulness for designing the rehabilitation robotic device for physically weak persons.