Plot-Rice v1.0: A global plot-based rice benchmark dataset with spatiotemporal heterogeneity for scientific deep learning
Satellite platforms have become the principal data source for large-scale rice mapping. Numerous studies investigated the advantages of integrating deep learning models with satellite data for rice mapping, but these studies are typically confined to specific regions and the quality of customized da...
Saved in:
| Main Authors: | , , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-06-01
|
| Series: | International Journal of Applied Earth Observations and Geoinformation |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S156984322500216X |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Satellite platforms have become the principal data source for large-scale rice mapping. Numerous studies investigated the advantages of integrating deep learning models with satellite data for rice mapping, but these studies are typically confined to specific regions and the quality of customized datasets varies. The absence of a global, standardized satellite dataset for rice mapping benchmarking has long resulted in both substantial redundant data processing efforts and challenges in evaluating new algorithms under a unified benchmark. To address these deficiencies, this paper introduces Plot-Rice v1.0, a global heterogeneous rice benchmark dataset based on Sentinel-1 and Sentinel-2 images, along with an automated framework that integrates SAR temporal features and leverages the foundation image segmentation model SAM-2 to generate plot-level rice samples. This framework effectively addresses the limitation of SAM-2, which supports zero-shot segmentation but lacks classification capabilities. Plot-Rice comprehensively accounts for the spatiotemporal heterogeneity of rice, incorporating plot-level rice labels and corresponding multi-source feature time series from 20 countries globally. In terms of feature composition, Plot-Rice integrates multi-source features, including radar backscattering coefficients, normalized backscattering coefficients, radar vegetation indices, spectral indices, and spectral combinations, constructing time series for 12 months of 2023 to facilitate temporal analysis. Finally, using this dataset, a series of benchmark experiments is conducted, encompassing performance comparisons of models such as U-Net, Res U-Net, Deeplabv3+, and HR-Net under multi-source feature time series, and comparisons of rice distribution extraction performance across diverse countries. Several recommendations are summarized for artificial intelligence (AI) in rice mapping to serve as a reference, contributing to global food security solutions. |
|---|---|
| ISSN: | 1569-8432 |