Edge-Aware Crop Field Boundary Mapping From Multimodal Sentinel-1 and Sentinel-2 Data
IEEE Geoscience and Remote Sensing Letters · 2026 · 23 · 1-5
DOI: 10.1109/lgrs.2026.3673620 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Geoscience and Remote Sensing Letters, 2026. Verfasst von Mehmet Furkan Celik und 5 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 15 Literaturangaben. Gefördert von: Food and Agriculture Organization of the United Nations (FAO) through the project “Developing a Geodatabase to Support Agricultural Planning” funded by Japan t…, FAO through the Swiss Federal Office of Agriculture.
Abstract (Originalsprache)
Crop field boundaries play a pivotal role in agricultural monitoring. Accurately mapping these boundaries using satellite data, however, remains a significant challenge as the quality of the results is strongly influenced by the field size and the phenological state of the study area. Leveraging the complementary information of Sentinel-1 (S1) and Sentinel-2 (S2) data, this work proposes an edge-aware crop boundary mapping approach that fuses multitemporal S1 with S2 monthly composites in a dual-stream U-Net supported by spatial and channel squeeze-and-excitation (scSE) and mid-level feature fusion. To refine the boundary delineation, we introduce a composite loss combining binary cross-entropy (BCE), Tversky loss, and a Sobel edge penalty. Our experiments demonstrate that the proposed multimodal U-Net architecture achieves the highest accuracy, with the Sobel loss term contributing edge sharpness and continuity. On the publicly available AI4SmallFarms benchmark, which includes over 400 000 field polygons across 62 tiles ( $5{\,}\times {\,}5$ km) in Vietnam and Cambodia, our approach produces sharper, continuous boundaries and fewer spurious edges, particularly for smallholder farms with highly fragmented agricultural areas. The benchmark is extended by adding analysis-ready, multitemporal S1 VH images (can be downloaded from https://doi.org/10.5281/zenodo.18921642) coregistered with the existing S2 data. The source code for the proposed framework is available at https://github.com/mf-celik/multimodal-crop-boundary
Förderung
- Food and Agriculture Organization of the United Nations (FAO) through the project “Developing a Geodatabase to Support Agricultural Planning” funded by Japan t…
- FAO through the Swiss Federal Office of Agriculture
Zitieren
APA
Celik, M. F., Persello, C., Nelson, A., Conchedda, G., Tubiello, F., & Paris, C. (2026). Edge-Aware Crop Field Boundary Mapping From Multimodal Sentinel-1 and Sentinel-2 Data. IEEE Geoscience and Remote Sensing Letters, 23, 1-5. https://doi.org/10.1109/lgrs.2026.3673620
Vancouver
Celik MF, Persello C, Nelson A, Conchedda G, Tubiello F, Paris C. Edge-Aware Crop Field Boundary Mapping From Multimodal Sentinel-1 and Sentinel-2 Data. IEEE Geoscience and Remote Sensing Letters. 2026;23:1-5. doi:10.1109/lgrs.2026.3673620
BibTeX
@article{Celik2026,
author = {Celik, Mehmet Furkan and Persello, Claudio and Nelson, Andrew and Conchedda, Giulia and Tubiello, Francesco and Paris, Claudia},
title = {Edge-Aware Crop Field Boundary Mapping From Multimodal Sentinel-1 and Sentinel-2 Data},
journal = {IEEE Geoscience and Remote Sensing Letters},
year = {2026},
volume = {23},
pages = {1--5},
doi = {10.1109/lgrs.2026.3673620}
}
EUVIMED