Non-stationary signal classification using the undecimated wavelet packet transform

Abstract

A classifier for non-stationary signals is presented in this paper. A time-frequency signal representation is calculated using the undecimated wavelet packet transform. The classification is performed with a support vector machine. Only the highest valued wavelet coefficients are selected as features in order to reduce the effect of noise. This classifier is compared against a classifier using a Wigner-Ville representation on a wideband non-stationary signal. The classifier based on the undecimated wavelet transform achieved a higher classification accuracy. Using only the largest half of the wavelet coefficients increased the classification accuracy