Metodologia para classificação de sinais EMG para controle de próteses com baixo esforço computacional

Abstract

This work describes the development of a muscular contraction originated signal (EMG) analysis system. Study and development of extraction and classification methods for these signals were made, so that they cold be recognized by an upper limb prosthesis with four degrees of freedom. To achieve such results, EMG signals from the biceps and triceps were classified in four distinctive patterns: elbow flexion, elbow extension, wrist pronation and wrist supination. Those patterns were classified by an artificial neural network, which received as inputs the characteristics of the EMG signals, extracted through detection of activation times and integral below the envelope. Analysis were made considering five pairs of electrodes, two located on the bíceps (long head (B1) and short head (B2)), and three on the tríceps ( long head (T1), short head (T3) and medium head (T2)). Dynamic and static contractions were evaluated during the experiments. As most of existing techiques rely on computationally demanding algorithms and complex mathematic analysis, the goal of this work was to find a simple and compact method to execute such tasks with the same performance, by use of simpler and more functional computational techniques, when compared with other well-known methods which achieve good results.