Diseño y desarrollo de un sistema embebido de identificación de patrones de señal EMG basado en algoritmos de aprendizaje automático.

Abstract

This work is part of the “Mano Esperanza” research project. An embedded pattern recognition system has been implemented, using electromyography (EMG) signals detected in real time to control a robotic prosthesis. The work has been divided in two stages: the stage of extraction of EMG signals that consisted in the investigation and application of the distinct methods of acquisition and conditioning of EMG signals and the stage intended for pattern recognition, where neural network algorithms and supervised learning algorithms were investigated. In addition, an algorithm was designed and implemented to train and validate the system. The results obtained from the patterns classification in the validation stage are presented through confusion matrices and percentages of accuracy to show the effectivity of the system, obtaining a success rate of 97%. The validation tests have been accomplished on 12 people without motor disabilities. The results show that the system was able to control the execution of five movements of the robotic hand prosthesis.