Classificação de padrões inerciais e eletromiográficos para aplicação em estratégias de controle de próteses de membro superior

• Ano de defesa: 2015
• Autor(a) principal: Arantes, Ana Paula Bittar Britto
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Uberlândia; BR; Programa de Pós-graduação em Engenharia Biomédica; Engenharias; UFU
• Palavras-chave em Português: Prótese; Eletromiografia; Prótese mioelétrica de membro superior; Máquina de vetor suporte; Reconhecimento de padrões; Sensores inerciais; Electromyography; Myoelectric upper limb prosthesis; Support vector machine; Pattern recognition; Inertial sensors; CNPQ::ENGENHARIAS::ENGENHARIA BIOMEDICA
• Link de acesso: https://repositorio.ufu.br/handle/123456789/14098; https://doi.org/10.14393/ufu.di.2015.383

In Brazil there is an expressive number of people with some kind of amputation, caused by accidents, or congenital. The technology, in this context, could be a great ally to help these people to recover their autonomy and further daily tasks, besides improviding their inclusion in society. People with this disability, mainly due to accidents, have difficulty in adapting to the lack of member; they are also psychological affected. For this reason, many studies have been made in development of artificial members that are increasingly functionals and intuitive to facilitate user's adjustment with the devices. The prosthesis that presents higher functionality and has been clearly accepted for the users are the myoelectric prostheses. This type of prostheses uses electromyography signal features to generate commands responsible for it movements. However, the system of control are quite complex and need to be increasingly improved. Many studies are focused only in control strategies that is more effective for this purpose. The objective of this research is to classify inertial and electromyographic patterns to apply in control strategies of upper member prostheses. For this, it was collected signals from two groups of people using four types of sensors: accelerometer, gyroscope, magnetometer and electromyography. The inertials sensors were added to the system for assessing patterns generated from small movements of the muscles. Groups of people were divided into healthy people, composed of seven individuals and people with unilateral amputation, composed of five individuals. Posteriorly it was made the extraction of sixteen characteristics of the collected signals and this signals was classified using an SVM classifier (support vector machine). Successively it was made the statistical evaluations as precision, sensitivity, accuracy and specificity of each class beyond the calculation of the average classifier accuracy rate. The results were promising regarding the use of inertial sensors. The group of healthy person, five had values of accuracy of statistical variables, specificity, accuracy and sensitivity greater than 95% and two above 90%. Regarding the group of people with amputations, three out of the five amputees showed statistical values above 90% . Along sixteen characteristics evaluated, those which are most relevant to reach these results were the fuzzy entropy, interquartile, peak amplitude, average absolute value of the second derivative and the average frequency of the signal. Furthermore, the results were more promising when EMG sensors and inertial characteristics were extracted jointly, reaching values above 90% of average hit rate of collected signals of both the stump as the intact arm using only inertial sensors for four out of the five people the amputee group.