Comparação eletromiográfica de métodos de entropia durantecontrações dinâmicas: fadiga neuromuscular

• Ano de defesa: 2024
• Autor(a) principal: Lex, Ariely Bertolani
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Tecnológica Federal do Paraná; Cornelio Procopio; Brasil; Programa de Pós-Graduação em Engenharia Elétrica; UTFPR
• Palavras-chave em Português: Entropia; Sistemas de reconhecimento de padrões; Redes neurais (Computação); Entropy; Pattern recognition systems; Neural networks (Computer science); CNPQ::ENGENHARIAS; Engenharia Elétrica
• Link de acesso: http://repositorio.utfpr.edu.br/jspui/handle/1/36411

Signal processing methods and machine learning can be applied in the biomedical field based on biological signals. One example of such signals is the electromyography (EMG) signal, which has applications in sports medicine and physical rehabilitation. This signal can be used to analyze neuromuscular fatigue, defined as the decline in muscle strength during muscle contraction. This study aims to compare the performance of fatigue analysis techniques in dynamic contraction exercises using entropy measurements, as well as to employ a predictive model based on a one-dimensional convolutional neural network to analyze features extracted from the signals. Initially, by assessing data separability and extracting features using Hjorth parameters, the goal is to classify fatigue and the absence of fatigue within the dataset. The dataset consists of signals from three quadriceps femoris muscles of the lower limb, recorded from 19 volunteers during a pedaling activity. Regarding Spectral Entropy and Shannon Entropy measurements, the muscle with the best performance in terms of angular coefficient was the rectus femoris (RF) muscle. The Convolutional Neural Network (CNN) achieved an accuracy of 61.78%. It is concluded that fatigue analysis methods based on median frequency and entropy measurements demonstrate good effectiveness in pattern identification, contributing to the understanding of changes in electromyography (EMG) signals. However, the application of the CNN model yielded unsatisfactory results, indicating that further adjustments are necessary to improve performance.