|
|
|
||||||||||||
|
|||||||||||||||
|
|
| Remote Sensing Meta Modal Representation For Missing Modality Land Cover Mapping: From EarthMiss Dataset To MetaRS Method | ||||
| Code | Dataset | Paper | ||||
|
1.Description
The EarthMiss dataset is a large-scale multi-modal benchmark designed to advance all-weather and multi-source land-cover mapping by integrating optical and Synthetic Aperture Radar (SAR) imagery. It provides globally distributed, co-registered image pairs with fine-grained semantic annotations, enabling robust analysis of heterogeneous land surfaces under diverse environmental and sensing conditions. The EarthMiss dataset contains 3,353 pairs of SAR-Optical image tiles with a spatial resolution of 0.6 m, covering 13 representative cities across five continents. Each sample is annotated with 8 semantic land-cover classes, including building, road, water, bare land, forest, farmland, playground, and background, ensuring complete coverage of both urban and natural environments. High-quality pixel-level semantic masks are generated through expert manual annotation and cross-modal alignment. This dataset establishes a unified framework for multi-modal fusion and modality-missing learning, supporting evaluation under full-modality, single-modality, and missing-modality conditions. The EarthMiss benchmark therefore provides a comprehensive foundation for developing and validating robust models for high-resolution, all-weather, and cross-regional land-cover mapping.
2.Annotation format Semantic category labels: background-1, building-2, road-3, water-4, barren-5,forest-6, agriculture-7, playground-8. The no-data regions are assigned 0. The color map is as follows:
Background = (255, 255, 255),
Building = (255, 0, 0),
Road = (255, 255, 0),
Water = (0, 0, 255),
Barren = (159, 129, 183),
Forest = (0, 255, 0),
Agricultural = (255, 195, 128),
Playground = (165,0,165)
3.Download
We hope that the release of the EarthMiss dataset can promote the development of all weather and multi-source land-cover mapping. You can click the link below to download the data. The city-wise data split is shown in Fig. 2. ● Baidu Drive: download, password: jfsq ● Zenodo: download
4.Copyright The copyright belongs to Intelligent Data Extraction, Analysis and Applications of Remote Sensing(RSIDEA) academic research group, State Key Laboratory of Information Engineering in Surveying, Mapping, and Remote Sensing (LIESMARS), Wuhan University, China. The EarthMiss dataset only can be used for academic purposes and need to cite the following paper, but any commercial use is prohibited. Any form of secondary development, including annotation work, is strictly prohibited for this dataset. Otherwise, RSIDEA of Wuhan University reserves the right to pursue legal responsibility. [1] Y. Zhou, A. Ma, J. Wang, Z. Chen, and Y. Zhong, "Remote sensing meta modal representation for missing modality land cover mapping: From EarthMiss dataset to MetaRS method," Remote Sensing of Environment, vol. 333, p. 115132, 2026 5.Contact If you have any the problem or feedback in using EarthMiss dataset, please contact: Mr. Yiheng Zhou: zhouyiheng@whu.edu.cn Prof. Ailong Ma: maailong007@whu.edu.cn Prof. Yanfei Zhong: zhongyanfei@whu.edu.cn |
|
