Verification of A Fast Training Algorithm for Multi-Channel sEMG Classification Systems to Decode Hand Configuration

Abstract

In this study, we evaluated a fast training algorithm to decode human hand configuration from sEMG signals on the forearms of five subjects. Eight skin surface electrodes were placed on the forearm of each subject to detect the sEMG signals corresponding to four different hand configurations and relax state. The preamplifier, which has 100 - 10000 times amplification gain and a 15 - 500 Hz bandpass filter, was designed to amplify the signals and eliminate noise. In order to enhance the performance of the classifier, feature extraction using class information was developed. The randomly assigned non-update learning method guarantees high speed classifier learning. The algorithm has been verfied by experiments with five subjects.