Developing a Controlling System Based on Electromyography (EMG) Signal Classification Using Deep Learning Models

Abstract

Recently, with the continuous improvement of control and influence technology, there has been an urgent need for people with disabilities for human-computer interface systems based on Electromyography (EMG) signals. When muscle cells are electrically or neuralgically triggered, the EM G signal measures the electrical potential that the cells produce. Human movement can be identified and detected by analyzing EMG signals. In this study, we propose an effective EMG-based controlling method. This system can classify EMG signals for different human muscle movements such as arm up, arm down, etc. We developed our own EMG signals dataset benchmark, which is measured using the MYO Armband device. The proposed methodology for signal classification is as follows. Firstly, the 8-channel EMG data of the forearm was obtained by the device. Comprehensive preprocessing techniques, including band-pass filtering, noise reduction, and signal standardization, were applied to enhance data quality. Then distinctive features were extracted to classify and identify the gestures for eight classes: open hand, closed hand, arm Up, arm Down, Ok sign, point to sign, victory sign, and wave (bye-bye sign). In the classification phase, different machine learning classification models were implemented such as Linear Discriminant Analysis (LDA), Decision Tree, Support Vector Machine (SVM) and Random Forest. Also, deep learning models were examined as Convolution Neural Network (CNN), Inception Network, Alex Network, and VGG Network which achieved the best average accuracy of 98%. Finally, these movements are used to send commands to control a Music playlist system.