A.V. Getmanchuk - Head of Sector, JSC «Taganrog Scientific Research Institute of Communications» E-mail: niis@pbox.ttn/ru

In article the problem of objects classification in the conditions of parametrical uncertainty and class intersection is considered. The given problem is solved in the absence of reliable information about parameters of a priori known classes presented in the catalogue of reference values in the form of confidence intervals of signs. The review of existing methods of objects classification for their application for the task in view solution is made. G.V. Sheleykhovskiy\'s method of objects classification, that takes into account required restrictions and possessing high degree of reliability of classification is considered. The basic imperfections of the method complicating its application in real conditions are specified. One of the basic imperfections of the method is the convergence problem. The combined method of classification based on G.V. Sheleykhovskiy\'s method is developed. The new method possesses the improved productivity and is devoid of known imperfections of the basic method. The developed combined method of classification uses the central idea of G.V. Sheleykhovskiy-s method - this is application of procedure of consecutive rationing, it allows to maximize entropy of system and to receive the least questionable distribution of probabilities. Also the method applies parametric space estimating a relative position of classification objects and classes from the catalogue of reference values. Laws of influence of classification objects on distribution of probabilities inside group of simultaneously observable objects are revealed. It has allowed to find ways of reduction of computing labor content of the method. Such approach has solved a convergence problem and this problem is solved without utilization of additional computing efforts. In article the developed algorithms that realize a new method are considered in detail. Experimental researches which have been made within the scope of the present work are described. The received results are shown. The present results show advantage of the new combined method of classification over G.V. Sheleykhovskiy\'s method known before.

 

1. Dzh. Tu, R. Gonsales. Principy raspoznavanija obrazov // M.: Mir. 1978.
2. ZHuravlev JU.I. Ob algebraicheskom podkhode k resheniju zadach raspoznavanija i klassifikacii // Problemy kibernetiki. M.: Nauka. № 33. 1978.
3. Dorogov A.JU. Teorija i proektirovanie bystrykh perestraivaemykh preobrazovatelejj i slabosvjazannykh nejjronnykh setejj. SPb.: Politekhnika. 2014.
4. Tehrano T., Asai K., Sugehno M. Prikladnye nechetkie sistemy. M.: Mir. 1993.
5. Kruchinin I.I., Loginov B.M. Modelirovanie processov klassifikacii mnogomernykh obektov peresekajushhikhsja klassov na osnove predstavlenijj nechetkikh mnozhestv // Trudy 2-jj Rossijjskojj konf. molodykh uchenykh po matematicheskomu modelirovaniju. M.: MGTU. 2002.
6. Kruchinin I.I., Loginov B.M. Modelirovanie processov klassifikacii mnogomernykh obektov peresekajushhikhsja klassov na osnove gibridnykh nejjrosetejj // Trudy 2-jj Rossijjskojj konf. molodykh uchenykh po matematicheskomu modelirovaniju. M.: MGTU. 2002.
7. Krug P.G. Nejjronnye seti i nejjrokompjutery: Ucheb. posobie po kursu «Mikroprocessory». M.: Izd-vo MEHI. 2002.
8. Bregman L.M. Dokazatelstvo skhodimosti metoda SHelejjkhovskogo dlja zadachi s transportnymi ogranichenijami // L.M. Bregman. Vychislitelnaja matematika i matematicheskaja fizika. M.: 1967.
9. Grishkov A.F., Kulakov A.A., SHpak V.F. Klassifikacija obrazov radiotekhnicheskikh signalov // Voprosy specialnojj radioehlektroniki. 2009. № 2.
10. Trukhaev R.I. Metody prinjatija reshenijj v uslovijakh neopredelennostejj. M.: Nauka. 1981.
11. Jaynes E.T. Information Theory and Statistical Mechanics // Physical Review. 1957. 106 (4): 620−630. DOI:10.1103/PhysRev. 106.620. Bibcode: 1957PhRv..106..620J.
12. Jaynes E.T.  Information Theory and Statistical Mechanics II // Physical Review. 1957. 108 (2): 171−190. DOI:10.1103/PhysRev. 108.171. Bibcode: 1957PhRv..108..171J.
13. SHpak B.F., Getmanchuk A.V., Radchenko S.A., Kulakov Andrejj A. K voprosu o skhodimosti algoritma klassifikacii radiotekhnicheskikh obrazov po metodu G.V. SHelejjkhovskogo // Voprosy specialnojj radioehlektroniki. Taganrog: FGUP TNIIS. 2011. № 2.
14. Sinkhorn Richard. A relationship between arbitrary positive matrices and doubly stochastic matrices. Ann. Math. Statist. 1964.
15. Sinkhorn Richard, & Knopp Paul. Concerning nonnegative matrices and doubly stochastic matrices. Pacific J. Math. 1967.