Recognition of Hand Grasp Pattern via EMG Signals Using Neural Network Classifier

Abstract

The aim of this study is the easy and efficient classification of five hand grasp pattern. For this aim, biomedical signals were used, which recorded via two-channel EMG sensors placed on the forearm muscles. At first stage of the study, a series of pre-processing steps (filtering and separating) was applied to the signals, then these signals were subjected to the feature extraction process. In this process, three-time domain and two frequency domain features were calculated for each channel. Finally grasp type was recognized using Neural Network (NN) classifier. Average classification success rate calculated as 96.40%. Based on this high success rate, it is said that bioelectrical based cognitive interaction method used in the study are suitable for prosthetic and orthotic devices control.

Keywords: sEMG, Pattern recognition, Grasp