A Novel Deep Learning Analysis in Detecting Urban Flood Objects for Faster Rescue Operations
IEEE Access · 2026 · 14 · 97856-97879
DOI: 10.1109/access.2026.3706444 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Padmavathi Lambu und 1 weiteren Autor:innen. Die Arbeit nennt 52 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: VIT-AP University under the V-Launch Internal Seed Grant.
Abstract (Originalsprache)
The impact of urban floods is increasing due to global warming. Flash floods caused by cloudbursts are particularly severe, creating significant havoc and disrupting human lives. Current early warning systems primarily concentrate on river flooding activities, including inundation areas, flood level indicators, and flood velocity calculations. But the early warning systems related to urban floods affected objects like trees, vehicles, drainage pits, and humans due to heavy water accumulation in urban areas. Limited research took place in the identification of urban flood-affected objects instantly during flash floods and heavy rains, which increases flood impact and leads to major devastation in urban areas. The current machine learning and deep learning algorithms identify ground-based objects but achieve limited classification accuracy. The current work built a custom urban flood dataset that comprises 15 different classes and 7320 images. Especially this work considered the identification of six prominent types of urban flood-affected objects, such as fallen trees, fallen electricity poles, manholes, floating or submerged vehicles, immersed people, and water-level markers. The proposed work conducted a technical performance analysis of the latest deep learning methodologies of YOLO models, which identified urban flood impact objects at higher classification accuracy. This work used four different versions of YOLO, which include YOLOv8 to YOLOv12, and obtained a finer classification of 96.4%, 97.4%, 95.8% and 91.9% on a custom urban flood dataset for YOLOv8, YOLOv9, YOLOv11 and YOLOv12, respectively. From the object classification results, it is observed that YOLOv9 performance is higher in classifying and detecting flood objects, and it helps for faster rescue operations.
Förderung
- VIT-AP University under the V-Launch Internal Seed Grant
Zitieren
APA
Lambu, P., & Duvvuru, R. (2026). A Novel Deep Learning Analysis in Detecting Urban Flood Objects for Faster Rescue Operations. IEEE Access, 14, 97856-97879. https://doi.org/10.1109/access.2026.3706444
Vancouver
Lambu P, Duvvuru R. A Novel Deep Learning Analysis in Detecting Urban Flood Objects for Faster Rescue Operations. IEEE Access. 2026;14:97856-97879. doi:10.1109/access.2026.3706444
BibTeX
@article{Lambu2026,
author = {Lambu, Padmavathi and Duvvuru, Rajesh},
title = {A Novel Deep Learning Analysis in Detecting Urban Flood Objects for Faster Rescue Operations},
journal = {IEEE Access},
year = {2026},
volume = {14},
pages = {97856--97879},
doi = {10.1109/access.2026.3706444}
}
EUVIMED