Hand movement classification using transient state analysis of surface multichannel EMG signal

Abstract

This paper presents two methods for the classification of six different hand motions based on the analysis of the transient state of surface multichannel electromyographic signals recorded from 10 normally limbed subjects. The signals were classified using the coefficients extracted from a discrete wavelet transform analysis. While the first method uses a feature vector based on the variance of the wavelet coefficients, the second analysis considers a PCA treatment focused on dimensionality reduction. These vectors were used to feed an artificial neural network. The first method was applied for both the transient and steady states obtaining an average classification accuracy of 89.43% (SD 2.05%) and 91.86% (SD 3.17%) respectively. The second method gave a classification accuracy of 92.58% (SD 3.07%) for the transient state. This proves the existence of deterministic information within the transient state of the EMG signal and the possibility to classify different movements since the beginning of the muscle contraction.