Comparative study of myoelectric pattern recognition based on wavelet analysis

Abstract

The choice of an appropriate wavelet family with a fast and robust classifier is an important step in the construction of a myoelectric control pattern recognition system for a prosthetic hand. In this study, five hand motions were classified using six wavelet functions and employing two different recognition methods. The selected classifiers are the support vector machine (SVM) and probabilistic regression neural network (PNN). The selected wavelet families used to decompose the recorded surface electromyographic (sEMG) signals are Biorthogonal (bior), Coiflet (coif), Daubechies (db), and Symmlet (sym). The results of our experiment demonstrate that the use of wavelet families at a high decomposition level increases the recognition rate of hand motions. The highest classification rate achieved was 96%, which was accomplished using the PNN classifier based on coif4 at the sixth decomposition level.