UDC 004.932
NEURAL NETWORK FLOOD DETECTION TECHNOLOGIES BASED ON DATA FROM SENTINEL-2 SPACECRAFT
A. S. Vendin, Student, R&D Technician at Research Institute «Photon», RSREU, Ryazan, Russia; orcid.org/0009-0005-1379-5580, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
S. A. Laryukov, Post-graduate student at RSREU, Junior Researcher at Research Institute «Photon» RSREU, Ryazan, Russia;
orcid.org/0009-0009-9082-1454, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
The problem of flood segmentation and water body detection based on satellite imagery from Sentinel-2 mission is considered. The aim of this work is to develop automated software tools that allow for automatic and accurate monitoring of water bodies using remote sensing data and artificial intelligence methods. The paper analyzes known approaches to solving this problem and identifies their shortcomings, which explains the relevance of new research in this area. Due to limited applicability of existing methods, as well as low quality of labeling in public datasets, a neural network based on «Lanky U-Net» architecture and trained on manually annotated dataset is proposed. As a result of the research, the following objectives were ad dressed: preparation and labeling of multispectral images obtained from Sentinel-2 satellites showing flood ed areas in Russian regions; training a model for water body detection; evaluating model performance using accuracy, precision, recall, and loss values; and flood detection using a static water body mask. Numerical indicators of segmentation quality and experimental results are presented, demonstrating the effectiveness of the proposed approach.
Key words: convolutional neural networks, machine learning, water bodies segmentation, automatic flood detection, Earth remote sensing data processing.