Indian Journal of Science and Technology

Abstract

Objectives: Study of sEMG signals of the hand gestures is important in designing hand prosthesis. Designing a sEMG pattern recognition system to control a myoelectric hand using neural networks is the objective of this study. Methods/ Statictical Analysis: The sEMG signal was acquired from flexor digitorumsuperficialis and extensor digitorum muscle of the ten healthy subjects by performing 12 different hand gestures. Six parametric feature extraction algorithms were applied to derive the prominent information from sEMG such as AR (Autoregressive) Burg, AR Yule Walker, AR Covariance, AR Modified Covariance, Levinson Durbin Recursion and Linear Prediction Coefficient. Recognition of the 12 gestures is accomplished using General Regression Neural Network, Probabilistic Neural Network and Radial Basis Function Neural Network. Findings: From the empirical results it was observed that the AR Burg and RBFNN combination had the highest recognition accuracy rate of 94.04%. Investigation also proved that recognition accuracy of sEMG signals were better for females when compare to males. It was also observed from the results that subjects in the age of 26-30 years had better muscle flexion compared to the other age groups studied. Application/ Improvements: In this paper the feasibility of recognizing twelve hand gestures from the sEMG of the hand and finger movements using neural networks was investigated. Six feature extraction algorithms and three neural networks were used to design algorithms for recognizing the twelve hand gestures
Keywords: Surface Electromyography, Autoregressive, AR Burg, AR Yule Walker, AR Covariance, AR Modified Covariance, Levinson Durbin Recursion, Linear Prediction Coefficient, Radial Basis Function Neural Network, Probabilistic Neural Network, General Regression Neural Network