Implementación en hardware de un clasificador de señales electromiográficas de miembro superior

Abstract

This work presents the development and hardware implementation of an electromyographic (EMG) signal classification model, through the selection of a preprocessing method and a convolutional neural network (CNN) architecture, evaluated on both a PC and a Raspberry Pi 4. The selection of these elements responds to a proposed global performance function, which integrates software and hardware performance metrics such as accuracy, execution time, and CPU usage, allowing the identification of the preprocessing method and the parameters of the CNN that present the best results obtained. The EMG signal classifier for the upper limb is implemented in hardware and is validated by a PC that emulates a MYO armband, sending the signal of the selected gesture via Wi-Fi to the Raspberry Pi 4 for real-time gesture classification.