Analysis of wavelet features for myoelectric signal classification

The common feature for classifying Motor Unit Action Potentials (MUAPs) of intramuscular myoelectric (EMG) signals is the Euclidean distance between the MUAP waveforms. The main effect which decreases the classification performance is MUAP shimmer. In this paper we show the relation between MUAP shimmer and the wavelet coefficients using a multiresolution approach and propose a selection of wavelet coefficients for classification. Simulations show that selected wavelet coefficients are better features for classification than the Euclidean distance of MUAP waveforms in the time domain.