Vankina I.N., Kroshilin S.V., Kroshilin I.S. The possibilities of neural networks for solving import substitution problems for domestic enterprises. Regional problems of economic transformation. 2023. No. 11(157). Pp. 67-76. DOI: 10.26726 ...

Текст статьи на сайте журнала URL: http://www.rppe.ru/new/index.php/rppe/article/view/2303 (дата ообращения 10.01.2024)

Abstract

The purpose of this work is to study the methods of semantic segmentation of images and implement the author's approach to processing the resulting image (from an unmanned aerial vehicle) based on building the architecture of neural networks and metrics in terms of increasing the efficiency of domestic production in the context of import substitution. The method or methodology of the work. The research materials of scientists devoted to semantic segmentation of images, as well as publications characterizing import substitution strategies in modern Russian realities, are analyzed and summarized. U-Net, FPN and MobileNetV2 networks are used as the initial set of neural network architectures. Results. Based on the results of the study, the authors proposed a solution for constructing and evaluating the accuracy of a non-network image processing model from UAVs obtained through combinations and modifications of various architectures to solve problems of domestic production in the context of import substitution. The theoretical and practical significance lies in the possibility of using neural network architectures in modeling images of captured UAVs in Russian production to solve import substitution problems. The authors present a software implementation and a comparative analysis of the methods of semantic image segmentation used to select a more effective one for solving the tasks set.