| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 1.64 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
A amputação de um membro pode ter várias causas, nomeadamente acidentes de trabalho e viação, ferimentos por armas de fogo ou congelamento. A utilização de próteses difere consideravelmente em diferentes países. Este trabalho surgiu com o objetivo de se estudar o funcionamento das próteses mioelétricas.
De um voluntário do sexo feminino foram recolhidos os dados (sinais mioelétricos). Testaram-se dois movimentos: abrir e fechar a mão. Os dados foram obtidos através de elétrodos de superfície colocados sobre a pele do voluntário e ligados a um sistema eletrónico de aquisição adequado a sinais fisiológicos ("PowerLab/4st" da ADInstruments). Aos dados adquiridos foi posteriormente realizada a extração de características e a sua classificação.
Recorrendo ao software Matlab foi realizada a extração de ruído, a extração de características e a classificação dos sinais recolhidos.
Para a extração de ruído e extração de características do sinal foi utilizada a Wavelet Packet Toolbox que aplicava a Transformada Wavelet Packet.
Para a classificação foram estudadas e comparadas duas técnicas de classificação: Redes Neuronais e Máquinas de Vetores de Suporte. Para isso foram utilizadas a Neural Pattern Recognition Toolbox e Classification Learner Toolbox, respetivamente.
O classificador de Redes Neuronais utiliza um vetor de target (vetor constituído apenas por zeros e uns) para classificar o sinal. O classificador de Máquinas de Vetores de Suporte utiliza um modelo de classificação para classificar os sinais. Comparando os dois classificadores, a Máquina de Vetores de Suporte apresenta melhores resultados relativamente ao classificador das Redes Neuronais.
The amputation of a limb can have various causes, including work and traffic accidents, injuries by firearms or freezing. The use of prosthetics differs significantly in different countries. This work was meant to study the operation of myoelectric prostheses. Data (myoelectric signals) was taken from a female volunteer. Two movements were tested: open and close the hand. Data were obtained through surface electrodes placed on the skin of voluntary and linked to an electronic system suitable acquisition physiological signs (“PowerLab/4st” of ADInstruments). Using the Matlab software was performed the noise extraction, feature extraction and classification of the collected signals. For the noise and features extraction was used Wavelet Packet Toolbox that applied the Wavelet Packet Transform (WPT). For classification was studied and compared two classification techniques: Neural Networks and Support Vector Machines (SVM). For this was used Neural Pattern Recognition Toolbox and Classification Learner Toolbox, respectively. The Neural Network classifier uses a target vector (vector only consists of zeros and ones) to classify the signal. The Support Vector Machine classifier uses a classification model to classify the signs. Comparing the two classifiers, the Support Vector Machine performs better relative to the Neural Network classifier.
The amputation of a limb can have various causes, including work and traffic accidents, injuries by firearms or freezing. The use of prosthetics differs significantly in different countries. This work was meant to study the operation of myoelectric prostheses. Data (myoelectric signals) was taken from a female volunteer. Two movements were tested: open and close the hand. Data were obtained through surface electrodes placed on the skin of voluntary and linked to an electronic system suitable acquisition physiological signs (“PowerLab/4st” of ADInstruments). Using the Matlab software was performed the noise extraction, feature extraction and classification of the collected signals. For the noise and features extraction was used Wavelet Packet Toolbox that applied the Wavelet Packet Transform (WPT). For classification was studied and compared two classification techniques: Neural Networks and Support Vector Machines (SVM). For this was used Neural Pattern Recognition Toolbox and Classification Learner Toolbox, respectively. The Neural Network classifier uses a target vector (vector only consists of zeros and ones) to classify the signal. The Support Vector Machine classifier uses a classification model to classify the signs. Comparing the two classifiers, the Support Vector Machine performs better relative to the Neural Network classifier.
Description
Keywords
Elétrodos Eletromiografia Máquinas de Vetores de Suporte Próteses Mioelétricas Redes Neuronais Transformada Wavelet Packet