Characteristics of Rational Classifications in Game-Theoretic Algorithms of Pattern Recognition for Unmanned Vehicles

In pattern recognition, depending on how the training sample is represented, a theory associated with the standard representation of the training sample (information) and the structural theory are distinguished. In the first case it is convenient to set and describe objects that are independent of each other, objects that have a rigid, static nature. In the second case, when information is specified in a structural way, it is convenient to study, describe and recognize objects whose basis presents an invariable internal or external structure. This paper considers a fundamentally different case of setting learning information. It deals with objects and processes that are based on conflict. Such features are peculiar to the data sets that an unmanned vehicle receives while driving, which requires minute-by-minute formation and decision-making in the context of the current traffic situation. Naturally, it is necessary to involve game-theoretic models to describe conflicts. It is these issues, i.e., the systematic study of game-theoretic models in the problems of recognition and classification, that the work is devoted to. It is emphasized that the classification of recognizable objects, as some speculative system, must meet the requirements of sufficiency of grounds and coverage of the totality of existing and possible recognizable objects. The most important purpose of classification is to describe the properties of its classes and subclasses, types and subspecies of recognized objects, which makes it possible for it to be used for recognizing specific objects, which subjects encounter in certain areas of activity. The paper introduces the concept of stable classification, formulates and proves the theorems based on the introduced concept, the sufficiency of the conditions for the correctness of the algorithm of pattern recognition, using the game-theoretic approach and evaluation of the degree of stability of the classification underlying it. The generated theorems allow the most optimal synthesis of algorithms for constructing a regular mapping and a matched continuation family, which, in turn, makes it possible for us to take into account the relationships in the training sample written in the form of a game-theoretic training model.