Multichannel Surface Electromyography System for Prosthesis Control Using RNN Classifier

In this paper, efficiency of using RNN classifier and multichannel surface electromyography to implement a control system for a bionic prosthesis is considered. The object of the study is the upper third of the forearm and the most commonly used actions in everyday life: pronation, supination, flexion, extension, grip, release. Registration of surface electromyography signals was carried out using eight electrode systems fixed on the cuff. Based on twenty-two days of data, it is possible to classify actions with an average accuracy of 43 % for a classifier trained on the first day, and improve the accuracy of the classifier up to 71 % by adding a new set of training data each day.