Pattern Classification of Combined Motions Based on Muscle Synergy Theory

Abstract

This paper proposes a novel pattern classification method of user's motions to use as input signals for human-machine interfaces from electromyograms (EMGs) based on a muscle synergy theory. This method can represent combined motions (e.g. wrist flexion during hand grasping), which are not trained by a recurrent neural network in advance, by combinations of synergy patterns of EMG signals preprocessed by the network. With this method, since the combined motions (i.e. unlearned motions) can be classified through learning of single motions (such as hand grasping and wrist flexion) only, the number of motions could be increased without increasing of the number of learning samples and the learning times for controlling of the machines such as a prosthetic hand. Effectiveness of the proposed method is shown by the motion classification experiments and prosthetic hand control experiments. The results showed that 18 motions, which are 12 combined and 6 single ones, can be classified sufficiently through learning of 6 single motions only (average rate: 89.2 ± 6.33%), and the amputee could control of a prosthetic hand using single and combined motions at will.