The Application of EMG and Machine Learning in Human Machine Interface

Abstract

Myocontrol is the intuitive control of a neural prosthetic via the user's voluntary muscular activations detected in the form of surface EMG(electromyography) signals. This type of control is usually implemented by means of pattern recognition, which uses a set of training data to create a machine learning model that can decipher these muscular electrical activations. EMG myocontrol is now widely used in multiple fields such as clinical medicine and human-computer interfacing. However, the usefulness of myocontrol depends greatly on the accuracy of the collected EMG signal. Some of the factors that cause inaccuracy include the interference of adjacent muscle, the inherent instability of the signals acquired from the human body (e.g. unexpected changes in the sEMG caused by sweating, electrodes displacement, muscle fatigue or specific postures and motions of the body segments ). Today, researchers focus on finding problems and propose new concepts of models that could make improvements. This review examines the application of EMG myocontrol in hand gesture recognition and neural prostheses, the shortages of EMG and surface EMG(sEMG) signal, as well as the solutions to some current issues.