DEEP LEARNING FOR DETECTING WATER BODIES IN UAV IMAGERY OF THE BRAZILIAN PANTANAL
DOI:
https://doi.org/10.14393/RCG2610776023Keywords:
Remote Sensing, Artificial Intelligence, Image ClassificationAbstract
The Pantanal, the largest continuous flooded plain in the world, faces preservation challenges due to the seasonal flooding cycle and human interventions. To better understand and preserve this biome, monitoring systems are essential, and the use of remote sensing techniques combined with advanced machine learning emerges as a promising strategy. This study investigated deep learning models for water body segmentation in UAV (unmanned aerial vehicle) images of the Pantanal. The images were captured using the MAVIC 2 Air camera, with a spatial resolution of 3 cm. Deep learning models such as InterImage, DeepLabv3+, and SegFormer were compared to evaluate their segmentation capabilities. A protocol was established for evaluation, considering metrics such as Intersection over Union (IoU) and Dice. SegFormer showed the best results, with an IoU of 96.16%, Recall of 97.85%, Precision of 99.46%, and an F1 Score of 98.04%. Although DeepLabv3+ and InterImage presented lower metrics, they also demonstrated robust performance. All models produced satisfactory results, but some difficulties were observed in accurately identifying water bodies.
Downloads
Downloads
Published
Issue
Section
License
Copyright (c) 2025 Lucas Yuri Dutra de Oliveira, João Lucas Aparecido Rocha Paes, Maximilian Jaderson de Melo, Maxwell da Rosa Oliveira, Eveline Terra Bezerra, Ana Paula Marques Ramos, Jonathan Li, Geraldo Alves Damasceno Júnior, Wesley Nunes Gonçalves, José Marcato Junior

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
Autores que publicam nesta revista concordam com os seguintes termos: a) Autores mantém os direitos autorais e concedem à revista o direito de primeira publicação, com o trabalho licenciado sob a Creative Commons Atribuição-NãoComercial-SemDerivações 4.0 Internacional. b) Autores têm permissão e são estimulados a publicar e distribuir seu trabalho online (ex.: em repositórios institucionais ou na sua página pessoal), já que isso pode gerar alterações produtivas, bem como aumentar o impacto e a citação do trabalho publicado. c) Em virtude de aparecerem nesta revista de acesso público, os artigos são de uso gratuito, com atribuições próprias, em aplicações educacionais e não-comerciais.




