Hand Gesture Recognition using sEMG Signals Based on CNN

Abstract

At present, the gesture recognition of surface electromyography (sEMG) is widely used. The traditional machine learning method often has the problem of feature extraction and incomplete information, which leads to low recognition rate. In this paper, a convolution neural network (CNN) based on deep learning is proposed for surface electromyography gesture recognition. the MYO Armband is used to collect the data firstly, then the mean square value and moving average window method are used to obtain the effective signal segment, then filtered, normalized preprocessing, and the data is expanded by sliding window. then the data set is fed into the CNN. after convolution, pooling, gradient descent and dropout layer processing, the classification model of 8 gestures is obtained, and then the model is evaluated with test set data. The experimental results show that the average recognition rate of 8 kinds of gesture movements is 96%, which verifies the feasibility of the method.