EMG CLASSIFICATION FOR THE PROSTHETIC HAND USING NEURAL NETWORK

Abstract

Electromyography (EMG) is procedure by which we measure the electrical activity of the muscles responsible for contraction and relaxation and can be used to track the abnormalities of mobility. Lately pattern recognition through EMG has been widely used with the traditional machine learning and deep learning methodologies to control the upper limb prostheses. The primary of this paper is the classification of hand gestures by obtaining EMG signals obtained from MYO bracelet channeled with 8 medical grade electrodes that provide signal information during muscle contraction. To categorize the hand movements i.e., radial and ulnar deviation, extension and flexion of wrist, hand at rest, supervised model of Artificial neural network is trained with the help of window function. A neural network is used to generate the model, which has input, middle, and output layers. A dataset of 4.2 million was constructed with the help of electrodes from where signals ae extracted seven features from each of the eight electrodes. After feature extraction the ANN model is trained with traditional back propagation technique to identify the given classes.  This strategy resulted in a success rate of 90% when classifying EMG signals that are collected from the upper arm muscles. The saved model is then transferred to the GUI-based model of NN, which is easy to understand and one can predict the outcomes. The research concluded that artificial neural network can be used in EMG pattern recognition and can be used in various aspects such as human-computer interaction and rehabilitation training.