Sistema de Aquisição de Sinais Mioelétricos e Detecção de Movimentos dos Dedos Usando Wavelets e Redes Neurais Artificiais

Abstract

This paper describes the development of a system for myoelectric signal acquisition from a person’s forearm that also classifies finger movements from features obtained in the frequency domain. Three electromyography surface electrodes were employed, two for the signals and one for reference. They were connected to a system composed of circuits for acquisition, conditioning, digitalization, and processing of the signals. The interface with the electrodes was performed by an analog circuit based on an instrumentation amplifier in a feedback loop with the person’s body, along with a lowpass active filter composed of operational amplifiers. A commercial digital acquisition device performed the conversion from analog to digital signals. The signal processing was composed of a temporal windowing, wavelet transform decomposition, estimation of the energies and entropies of the wavelet coefficients, and classification by multilayer perceptron artificial neural networks. During the method evaluation, we tested variations of wavelet filter and time window lengths and several neural network topologies to identify the most significant parameters for signal classification. As a result, after selecting the best window and filter lengths, the tested topologies presented accuracy rates between 73% and 83% for the classification of the five types of finger movements.