A pilot study on the daily control capability of s-EMG prosthetic hands by amputees

TL;DR

This study evaluates the stability of s-EMG signal classification using SVMs for prosthetic hand control, demonstrating promising accuracy and repeatability in a limited experimental setting.

Contribution

It investigates the repeatability and stability of SVM-based s-EMG classification for prosthetic control using two feature extraction methods.

Findings

Average accuracy above 73% in worst-case scenarios

SVM performance remains stable over short periods

Two feature representations tested: MAV and WL

Abstract

Surface electromyography is a valid tool to gather muscular contraction signals from intact and amputated subjects. Electromyographic signals can be used to control prosthetic devices in a noninvasive way distinguishing the movements performed by the particular EMG electrodes activity. According to the literature, several algorithms have been used to control prosthetic hands through s-EMG signals. The main issue is to correctly classify the signals acquired as the movement actually performed. This work presents a study on the Support Vector Machine's performance in a short-time period, gained using two different feature representation (Mean Absolute Value and Waveform Length) of the sEMG signals. In particular, we paid close attention to the repeatability problem, that is the capability to achieve a stable and satisfactory level of accuracy in repeated experiments. Results on a limited setting are encouraging, as they show an average accuracy above 73% even in the worst case scenario.