Tiny Object Detection in Super-Resolved Sentinel-2 Imagery

The detection of tiny objects in satellite imagery is a critical task with wide-ranging applications, including environmental monitoring, urban planning, disaster response, and the surveillance of critical transport infrastructure. Sentinel-2 satellite data, characterized by providing rich spectral...

Full description

Saved in:
Bibliographic Details
Main Authors: C. Ayala, J. F. Amieva, P. Vega, R. Perko, S. Aleksandrowicz
Format: Article
Language:English
Published: Copernicus Publications 2025-05-01
Series:The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Online Access:https://isprs-archives.copernicus.org/articles/XLVIII-M-6-2025/61/2025/isprs-archives-XLVIII-M-6-2025-61-2025.pdf
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The detection of tiny objects in satellite imagery is a critical task with wide-ranging applications, including environmental monitoring, urban planning, disaster response, and the surveillance of critical transport infrastructure. Sentinel-2 satellite data, characterized by providing rich spectral information at a moderate spatial resolution (10–60m), poses significant challenges for the identification of small-scale features due to limited spatial detail and the effects of mixed pixels. This study investigates the potential of super-resolution techniques to enhance Sentinel-2 imagery for improved tiny object detection. A dataset was meticulously annotated to identify aircraft across diverse areas of interest, enabling rigorous evaluation using advanced methodologies. Detection was performed using a hybrid approach that combines a YOLOv8-based object detector and a vision-transformer-based object density estimator. The fusion of these complementary methods significantly reduces false positives, resulting in improvements in precision and F1 score. The findings underscore that super-resolved Sentinel-2 imagery offers a viable and cost-effective solution for detecting tiny objects, particularly in scenarios where access to high-resolution imagery is restricted or economically prohibitive.
ISSN:1682-1750
2194-9034