The task of classifying objects for systems with limited computing resources

Abstract

The features of the task of classifying images for systems with limited computing resources that are decomposed according to the reference model for the interaction of open systems are determined. Among the systems that implement the functions of video surveillance, a recent development system and network based on mobile, in particular, aircraft, occupy a special place, the feature of which, in particular, UAVs, is the limited computing resource available for implementing the functions of collection, processing and transmission ( video, photo) information. One of the important tasks that are solved when implementing the surveillance function (video) in UAV systems and networks is the classification of images, that is, assigning the observed object to one of a given number of groups. An approach is proposed to improve the quality, in terms of cost / effectiveness, of classification systems with limited computing resources, for example, recognition and classification systems deployed on aircraft. A similarity measure has been introduced, which is characterized by a vector with a dimension corresponding to the parameters of intelligent sensors with which information about the object is taken. A simplification of the classification procedure is proposed, based on the steps of folding the vector similarity criterion into a scalar when forming the proximity matrix; representing classification objects as nodes of a computer network with a centralized topology, followed by applying the proximity matrix in algorithms that have long been used for topological design of computer networks.  A simplified classification procedure may be useful in classifying objects of a limited set, such as that formed by passengers and personnel of domestic air transportation system, the metro, etc. Refs:8 titles.