Finger Movement Regression with Myoelectric Signal and Deep Neural Network

Abstract

Research on electromyographic (EMG) signals is intensively carried out to help disabled people to control prosthetic hands. Neural Networks have been widely used in research on the classification of finger movements using EMG. The study of a classification system generally still works on a limited number of movements, even though the human body, especially fingers, has a nearly unlimited combination of movements to help do daily activities. To overcome this, a proportional control system is needed. In its recent development, research on myoelectric control using EMG devices is still in a laboratory environment. Hence, the results obtained in a clinical setting are often different. However, along with technological developments, the emergence of affordable and wearable commercial EMG devices such as Myo Armband, has encouraged this study to develop control systems of prosthetic fingers using regression. One of many options available is neural networks that have been widely used in various fields. By estimating each joint with a different neural network, the result shows the predicted is fitted to the actual angle with R2 high as 99%.