Motion classification by neural network using muscle vibration for electric prosthetic hand control.

Abstract

There are more than 80,000 upper-limb amputees in Japan, and many of them desire to use an electric prosthetic hand. Myoelectric prostheses are one type of electric prostheses, but they are affected by skin impedance and cannot be used for long periods of time. In this study, we measured muscle vibration using piezoelectric wire sensors, which has both muscle tone and muscle protuberance characteristics that are not easily affected by skin impedance. The results of eight motion classifications using a neural network showed that the average discrimination rate was 80% or higher for all subjects. When the motion classification was performed using only rock-paper-scissors motion, subjects A and B showed an average discrimination rate of more than 95%. These results suggest the possibility of motion classification including complex finger actions.