Abstract: In the given work displaying is built of great number of appearances on the great number of vectors of errors of pattern recognition by the neuron network, which allows to link classification of appearances with the analysis of vectors in space of errors. A criterion allows grouping appearances, recognizing and comparing them. A vector criterion is formulated of closeness of signal images in space of errors. An algorithm is offered for transition from space of parameters of signal images in space of errors of pattern recognition. An optimum decided rule is built for classification of signal images of signals with the use of weighed criterion of closeness of recognizable signal images in space of errors of recognition. Authenticity of the received scientific results, conclusions and recommendations of this thesis work has been confirmed by the results of experimental researches of the developed universal system of intellectual data analysis, which solve the task of recognition of objects of the electro-optical images NEFFClass BGCGG (Basic Gradient Conjugate Gradient Genetic), conducted on the base of „Institute of the Applied Systems Analysis” NTUU „KPI”. The results received in work evidently demonstrate efficiency of the use of the developed models, methods and algorithms for the solution of tasks of recognition of signals.

Keywords: neural networks, basic method, method of accreditation, vector criterion

Link:

РАСПОЗНАВАНИЕ ОБЪЕКТОВ С ИСПОЛЬЗОВАНИЕМ КРИТЕРИЯ НА ОСНОВЕ ВЕКТОРНОЙ МЕРЫ БЛИЗОСТИ ОБРАЗОВ В ПРОСТРАНСТВЕ ОШИБОК (Pattern recognition using a criterion based on vector criterion of patterns proximity in the space of errors)

Петр Четырбок

http://www.foibg.com/ijitk/ijitk-vol08/ijitk08-02-p09.pdf