About the features of the wavelet spectrum as a flat image feature space for statistical pattern recognition methods

Abstract

The methods of identification with a given accuracy of the mass of objects of natural origin, with high intraclass variability of the visual, the proximity of the classes themselves are considered. In particular, the problems of rapid diagnosis of latent infection of grain by insects in bulk grain and legumes. Theoretical and experimental studies with a variety of statistical classifier that uses pattern recognition algorithm to construct the ellipsoid scattering in the feature space (with a normal distribution), and the degeneracy of the covariance matrix characteristics and features of the spectrum of the wavelet feature space for expansion, were held. Proposed use of the degeneracy of the covariance matrix of the wavelet spectrum of a flat image to improve recognition.