Hand Electromyography Circuit and Signals Classification Using Artificial Neural Network

Electromyography (EMG) is the study of electrical activity of muscles signals. This technique can be used for the control of prosthetic for amputees or for medical purposes in muscular disorders. Major challenge faced in this domain is high cost of the devices to control the prosthetic. In addition to the cost of the device, number of parameters used for classification is large for studies in this domain. In this study we propose a low cost circuit for EMG signal extraction. We used 4 channels of proposed EMG circuit to classify 6 different motion that includes individual finger motions and fist motion. Despite being low cost, our circuit provides the signals that can be classified with high accuracies comparable to other studies. For classification, we used artificial neural network with less number of parameters to achieve accuracies comparable to other studies using higher number of parameters. We collected data from 5 healthy subjects using our proposed circuit. Behavior of EMG signal varies from subject to subject depending upon different factors. We used six features from time and frequency domains, gave an accuracy of 98.8% and 96.8% for all combined subjects with two different algorithms and an average accuracy of 99% with standard deviation of 0.6 for all individual subjects.