An Interactive Training System Using a Class Partial Kullback-Leibler Information Measure for EMG-based Prosthetic Hand Control

Abstract

This paper proposes a novel interactive training system using a class partial Kullback-Leibler (KL) information measure. The proposed training system is capable of selecting effective motions based on the partial KL information theory, and can perform EMG signal control training for selected motions. In the experiments performed, a three-days training session using the proposed training system was conducted with three subjects. The results showed that the number of motions was gradually increased through training and the classification rates on the final day was 99:6 ± 0:4 [%] with a high level of accuracy. These outcomes indicated that the proposed system was effective for the EMG-based prosthetic hand control training.