Shevchenko

The methodology of selecting of classifiers’ features was described in this article; these classifiers were used in processing of diffraction image of complex spatial objects. The substantiation of the separation of the object class into subclasses in terms of information theory, as well as reasonable selection of object classes with the usage of ellipsoidal estimation were given. The criteria of qualitative selection of classifier’s feature vector, answering questions about the size of teaching selection, the cardinality of the set of classes, the essential dimension of feature vector, were developed. Evidence-based approach to the selection of informative features of classifier of signals, based on the concept of entropy of information theory and ellipsoidal estimation, were proposed.