A Study on the Application of One Dimension Convolutional Neural Network for Classification of Gestures from Surface Electromyography Data

Abstract

Myoelectric control systems are gaining popularity with the availability of commercial, low-cost, surface electromyography sensors. These systems can be used for gesture recognition which finds application in human-machine interfaces. The gestures are recognized using pattern recognition algorithms. Machine learning or deep learning techniques can be applied for the classification of gestures. In this paper, a user-specific 1-Dimensional Convolution Neural Network is proposed for the classification of Surface Electromyography data recorded using a commercially available surface electromyography recording device to perform offline classification of 5 hand gestures using limited data of less than 400 samples. An average accuracy of 82%±3% was achieved during the study after cross-validation of the data using 5-fold stratified cross-validation.