Real Time Identification of Human Forearm Kinematics from Surface EMG Signal Using Artificial Neural Network Models

Abstract

This work identifies the human forearm kinematics in real time by utilizing the surface electromyography (SEMG) signal using two different artificial neural network models. Here, the SEMG signals from biceps brachii muscle are captured using Ag-AgCl electrodes. Two time domain features Integrated EMG (IEMG) and number of zero crossing (ZC) are derived from the measured SEMG signals after segmenting the raw SEMG signal into 250 millisecond window. These two time domain features are used as the input signals for identifying the human forearm kinematics. Human forearm kinematics is determined using two neural network models, (1) Multi Layered Perceptron Neural Network (MLPNN) model and (2) Radial Basis Function Neural Network (RBFNN) model. The results obtained from the both models are compared in this work. The results indicate that the RBFNN model is giving better identification results with an average regression coefficient value of 0.756767 and 0.389113 for the identification of angular displacement and angular velocity respectively when compared with MLPNN model.