Deep Learning Approaches For EMG-Based Motion Recognition and Force Estimation

Abstract

 In this study, we study the use of deep learning methods, in this case Convolutional Neural Networks (CNNs) to analyze EMG signals in order to improve prosthetic performance in upper limb disabled patients. This study focuses on gripping actions which are an important part of prosthetic hand function. Using a curated database, we analyze EMG signals with MATLAB R2023b and convert them into grayscale images with palm movements using an algorithm proposed in this paper. We then categorize the generated images from the algorithm into eight different categories using CNNs and find very promising results. SSAE-f and CNN performed similarly (p = 0.55) and reduced the differences in accuracy among the subjects compared to conventional methods which indicates their robustness and reliability. The study suggests that deep learning is effective for EMG-based gesture detection and will help in human-computer interface and healthcare technology, which will be applicable to prosthetic devices