Remote Sensing Change Detection via Weak Temporal Supervision

TL;DR

This work tackles semantic change detection with limited pixel-level annotations by introducing weak temporal supervision: extending single-date remote sensing datasets with additional temporal observations and creating bi-temporal change examples from non-overlapping locations. It leverages balanced batch sampling, an object-aware sIoU-based change map, and iterative dataset refinement to train a bi-temporal change detector without new annotations. Extensive experiments on extended FLAIR and IAILD datasets demonstrate strong zero-shot and low-data performance, plus robust out-of-domain generalization and large-scale qualitative results over France. The approach offers an annotation-efficient, scalable pathway for building-scale change monitoring in both aerial and satellite imagery.

Abstract

Semantic change detection in remote sensing aims to identify land cover changes between bi-temporal image pairs. Progress in this area has been limited by the scarcity of annotated datasets, as pixel-level annotation is costly and time-consuming. To address this, recent methods leverage synthetic data or generate artificial change pairs, but out-of-domain generalization remains limited. In this work, we introduce a weak temporal supervision strategy that leverages additional temporal observations of existing single-temporal datasets, without requiring any new annotations. Specifically, we extend single-date remote sensing datasets with new observations acquired at different times and train a change detection model by assuming that real bi-temporal pairs mostly contain no change, while pairing images from different locations to generate change examples. To handle the inherent noise in these weak labels, we employ an object-aware change map generation and an iterative refinement process. We validate our approach on extended versions of the FLAIR and IAILD aerial datasets, achieving strong zero-shot and low-data regime performance across different benchmarks. Lastly, we showcase results over large areas in France, highlighting the scalability potential of our method.

Figures (17)

• Figure 1: Large-scale building change detection comparison on BD ORTHO ign2025bdortho data (Lille metropolitan area in France, approximately 55.3 km²). Availability of annotated datasets has always been a challenge for semantic change detection. To avoid this problem, we propose a novel weak temporal supervision strategy that leverages additional temporal observations of existing single-date annotated data. This allows us to train robust and scalable models, without requiring any new annotations. Left: map presenting the reference building changes (demolitions and constructions) derived from the French IGN's OCS GE data product ign2025ocsge. Middle: result of a Dual UNet trained with our methodology, which closely aligns with the reference. Right: output of a Dual UNet trained on FSC-180k benidir2025change, which produces numerous false positives.
• Figure 2: Visual examples. For each of the extended datasets, we show example triplets ($S_{t}$, $I_{t}$, $I_{t'}$) in this order from left to right, corresponding to the annotation mask, the original image, and the added acquisition respectively. Pairs may exhibit significant land cover changes (top row), or irrelevant changes due to shadows or seasonal variations (middle and bottom rows).
• Figure 3: XOR vs. sIoU for change map generation from image pairs, for the building change detection binary task. Top: a real pair with slight viewpoint variation—XOR falsely detects changes, while sIoU correctly shows none. Bottom: a fake pair with different buildings—XOR misses overlapping changes, sIoU correctly marks both. Only one building per image is labeled for clarity.
• Figure 4: Qualitative results on b-FLAIR-test. We compare the building change maps predicted by baseline methods and ours.
• Figure 5: Samples removed during iterative refinement. Triplets ($I_{t}$, $I_{t'}$, $\hat{M}$) removed during refinement exhibit real building changes or blurred sensitive areas, where the assumption $M=0$ does not hold. Such samples are correctly identified by the model and excluded from the training set on subsequent iterations.