Remote Sensing Change Detection via Weak Temporal Supervision

Abstract

Change detection in remote sensing requires aligned bi-temporal images and corresponding change maps. However, annotation is expensive, and temporal pairs are often not available. We propose a weakly supervised approach that builds change detection datasets from already existing single-temporal datasets by adding newly acquired images of the same locations. Using the original annotations as weak supervision, we derive automatically generated change masks without requiring manual labeling of the temporal pairs. We demonstrate that this strategy produces high-quality training datasets and achieves competitive results on remote-sensing benchmarks.

Teaser

Data

Building on the observation that annotation is significantly more expensive than data acquisition itself, we propose to extend existing single-temporal datasets with new, aligned temporal acquisitions, using their existing annotations as weak supervision for bi-temporal change detection. Thus, we extend three single-date datasets to bitemporal:

Acknowledgements

This work was funded by AID-DGA (l’Agence de l’Innovation de Défense aà la Direction Générale de l’Armement—Ministère des Armées), and was performed using HPC resources from GENCI-IDRIS (grants 2023-AD011011801R3, 2023-AD011012453R2, 2023-AD011012458R2) and from the “Mésocentre” computing center of CentraleSupélec and ENS Paris-Saclay supported by CNRS and Région Île-de-France. Centre Borelli is also with Université Paris Cité, SSA and INSERM. This work is additionally supported by RGC-GRF project 11309925, Mathematical Formalization of GIS. We thank Etienne Bourgeat, Fabien Poilane and Floryne Roche for their valuable assistance in data curation and annotation. This template was originally made by Phillip Isola and Richard Zhang for a colorful ECCV project; the code can be found here.