Monitoring of maritime navigation facilities based on aerospace remote sensing data in the microwave range using neural network technologies
M.L. Belokopytov, D.A. Shlyonskikh, S.V. Morozov, S.V. Sirota · Journal of Radio Electronics · 2022
The article is devoted to the issues of automated search for objects of maritime navigation on radar images. A comparative analysis of the Yolov5 neural network family has been carried out. To detect objects, a one-stage automatic detector was used, built on the basis of a convolutional neural network of the Yolov5x type and trained on the SAR Ship Dataset. Digital modeling of the proposed recognition system has been performed. Verification of the trained model was carried out, as well as evaluation of the quality of the convolutional neural network algorithm. The main difficulties encountered in the preparation of a training sample are considered. The ways of their solution are proposed. Conclusions are drawn regarding the possibility of using the developed detector in order to automate the process of recognition of marine objects.