Separation of Electromyography Signals for Hand Movements Using Selected Machine

Given the existing challenges in electromyography (EMG) signal processing, this paper presents a novel approach based on a combination of Short-term long-term memory classifiers, artificial neural network with linear discriminant analysis feature reduction (LDA), achieving results distinct from previous studies. Electromyography signals from hand flexor digitorum muscle of the forearm, were recorded using Adhesive and gel electrodes, And these signals are amplified by a SUBTLE chipset, and digitized using an Arduino Nano board and software. The ideal EMG signal is obtained after fourth-order Butterworth filtering. Key features were extracted in a ۱۵۰-second period without overlap. Results demonstrated that the Long Short-Term Memory (LSTM) classifier, with an accuracy of ۹۵.۸۳%, outperformed the Artificial Neural Network (ANN), which achieved ۸۲.۶۵% accuracy. The findings highlight an ۱۱.۲۲% superiority of deep learning in Electromyography signal processing, opening new horizons for the design of Smart prosthetics with natural control, robotic rehabilitation devices, developing prosthetic hands, and human-machine interface.

کلیدواژه ها:

Long Short-Term Memory (LSTM) units ، Artificial Neural Network (ANN) ، Deep Learning ، Motion Classification ، Electromyography