Desenvolvimento de um sistema de aquisição e classificação contínua de padrões em sinais mioelétricos baseada em wavelets e redes neurais artificiais

Abstract

This study describes the development of a myoelectric signal acquisition and classification system. Some results are presented to show the identification of patterns in these signals. The system contains: signal collection through surface electrodes for eletromyography, signal conditioning circuit, digitizing device and digital signal processing. The analog signal conditioning circuit is built on a printed circuit board, where amplification is performed by an instrumentation amplifier and closed-loop feedback, and the filtering is performed by an active filter circuit with operational amplifiers designed as a Chebyshev filter. This processing performs: first, the signal windowing at the moment the muscle contraction signal is identified, then the signal decomposition by Wavelet Transform and, at the end, the patterns classification of the coefficients in the Multilayer Perceptron Neural Network. In addition, is presented comparisons between different configurations in digital processing, in order to understand which parameters are most significant in classifying these signals.