Классификация и распознавание наземных объектов в потоке радиолокационных кадров на основе нейросетевого подхода в передней зоне обзора бортовых РЛС многопозиционной системы

Abstract

Listen

Translate article

Classification and recognition of ground objects in the stream of radar frames based on a neural network approach in the forward field of view of the onboard radar of a multi-position system. This article discusses the features of effective classification and recognition of ground objects in the video stream of radar frames formed in the forward field of view of the on-board radar of a multi- position system. The issues of building spatially distributed systems for on-board radar automated monitoring of the earth's surface are covered, modern requirements for the resolution of radar frames are determined, and the features of the formation of a video stream of radar frames for the implementation of a system for classifying and recognizing ground objects are discussed. To solve these problems, technical vision methods are used, in particular, radar frame segmentation for detecting, classifying and distinguishing ground objects against the background, as well as neural network methods implemented in the algorithms for complex processing of streaming data in the onboard multi-position system of aviation monitoring of the earth's surface. These methods allow, at the first stage of processing video frames, to quickly select each object into a class, while separating the static background in the video stream of radar frames. At the second stage, in order to recognize objects of the selected class, technical vision tools are used based on the use of multilayer neural networks. As a result, a method for classifying and recognizing ground objects in a stream of radar frames based on a neural network approach is described, and recommendations are given for their further practical use.