Redes neurais artificiais na classificação de movimentos de mão por sinais eletromiográficos

Prosthesis control through electromyographic signals has been used in many works, but it is still hard to achieve dexterity of a human hand. Current systems rely on the extraction of various features from the EMG signal to further classify user’s intention. In an attempt to contribute to the field, this paper investigates the performance of Multilayer Perceptron and Learning Vector Quantization to classify various classes of movement.  Electromyography data were collected from four pairs of electrodes positioned around the forearm of six subjects. Seven classes of movement were classified based on features extracted in the time domain. The results showed that both methods can achieve accuracy above 92%, with a slight advantage for the Multilayer Perceptron - Learning Vector Quantization: 92.69%; Multilayer Perceptron: 94.56%.