Investigation of Methods for Determining the Aspect Angle of a Spatially Distributed Target in its Radar Image

Abstract

The article is devoted to the problem of determining the aspect angle of a surface ship based on its radar image obtained in synthetic aperture radars. Among the main methods of solving this problem are the classical Bayesian method of multi-alternative hypothesis testing and its modifications, and/or the method of classifying surface ships located at different angles using artificial neural networks (ANNs). The study demonstrates that achieving high efficiency in aspect recognition using ANNs requires significant computational resources, as well as an access large, representative, and scalable training dataset. ANNs demonstrates high performance in various observation conditions in case sufficient computational and time resources; however, it is noteworthy that their effective training requires a substantial amount of processing time, reaching several hours. At the same time, classical methods are capable of performing calculations in fractions of a second, even on relatively low-powered devices. It is also worth noting that as the number of recognised classes increases, ANNs may consume up to tens of gigabytes of RAM, limiting the accessibility of this method in the aspect of task recognition of spatially distributed targets.