A Convolutional Neural Network-Based Architecture for EMG Signal Classification

Abstract

Electromyography (EMG) signals represent the electrical manifestation of neuromuscular activation and they contain valuable information on muscle activity. Recently, the use of EMG has increased in prostheses, rehabilitation, human-machine interaction applications, among others. Deep Learning (DL) methods showed effectiveness in the classification of EMG signals. In this paper, we propose a featureless approach using a Convolutional Neural Network (CNN) for EMG signal classification to identify different wrist and finger motions. The proposed approach does not need a set of features selected as an input. However, it automatically learns from the input EMG signals. Some simulation results are realized at the end of this work in order to evaluate the performance of the proposed CNN-based architecture. The used data set is the Ninapro Database 2, which includes 40 different subjects and 49 different motions.