Korobov A. Models and methods of information technology for autonomous video monitoring of terrain by an unmanned aerial vehicle.

Object of research ‒ the process of terrain autonomous monitoring by an unmanned vehicle under the limited resources and information restrictions; the aim of research ‒ with increasing a functional efficiency of the onboard system of an unmanned aircraft operating autonomously for visual navigation and recognition of objects of interest on the ground in conditions of incomplete certainty and resource constraints through the creation of information technology of machine learning; research methods ‒ methods of convolutional neural network, sparse coding methods, methods of coding theory and optimization; methods of information-extreme intellectual technology and technology of feedforward neural networks results ‒ models and methods of information technology for autonomous video monitoring of terrain by an unmanned aerial vehicle; novelty ‒ for the first time, a method of synthesis of information-extreme classification decision rules of an autonomous on-board unmanned aerial vehicle system is developed, which, unlike the known ones, is based on binary coding of a characteristic description of ensembles of decision trees with construction in the radial basis of a binary space of features optimal in the informational vertex, that make operative functioning and ensure reliable recognition of observations under the conditions of limited labeled sample; improved methods of machine learning extractor feature describe the environment by applying the principles of competitive learning, sparse coding, transfer learning and metaheuristic optimization that allows you to create the best in information and value terms feature description video surveillance for classification and regression analysis for a limited amount of labeled training data and computing resources; were further developed models calculations of feature extractor describe the environment through the application architecture convolutional networks, sparse coding and considering both local and contextual information that can improve the functional efficiency of detection of small objects of interest or obstacles in terms of resource and information constraints; the degree of implementation ‒ the result were implemented at the Missile troops and artillery research center, Sumy state university; industry ‒ intellectual information technology