Nizhebetska I. Method of normal transformation and its application for classification of signals by form

The goal of dissertation research is to increase the efficiency of signal classification by form by developing method of normal transformation. The object of research is the process of signal classification. The subject of research is the methods and algorithms for signal classification. In the dissertation work method and algorithms for creating discrete matrix operators of normal transformation are proposed for the first time. Main feature of normal transformation is that the spectrum of the selected reference signal of arbitrary form contains only one non-zero transform, while by weight of other components in the spectrum the degree of similarity/difference of the signal to reference can be calculated. A modified method of signal classification of maximum likelihood criteria, that requires no inversion of the correlation matrix of the training set and allows to perform the classification regardless the rank of correlation matrix, is offered. The method of normal transformation is applied to the task of authentification of person by the dynamically entered signature. The results of the method of signature components normal filtering is compared with the results of modified method of classification of the maximum likelihood critearia.