Kuro-Siwo-Webdataset
收藏资源简介:
Kuro Siwo webdatasets是一个用于快速洪水测绘的全球多时相合成孔径雷达(SAR)数据集。该数据集覆盖2015年至2022年全球六大洲和三种主要气候区发生的43次洪水事件,所有洪水区域标注均由专家团队通过细致影像解译完成,空间分辨率高达10米。针对每次洪水事件,数据集提供一景Sentinel-1卫星的灾后影像、两景灾前影像以及对应的数字高程模型(DEM)。数据集包含地距探测(GRD)和单视复(SLC)两种处理级别的SAR产品,并配套BlackBench基准测试所需的webdatasets,包括GRD和SLC产品的训练集与测试集。数据已预先划分为训练集和测试集,测试集覆盖全球不同环境条件的10次洪水事件以确保评估挑战性和代表性。在数据结构上,GRD数据集每个样本包含灾前和灾后影像的VV和VH极化波段(共6个波段文件)、DEM、样本元数据、洪水标注掩膜(0:非水体, 1:永久性水体, 2:洪水)及有效像素掩膜;SLC数据集每个样本包含灾前和灾后复数据图像、DEM、元数据及相同掩膜文件。该数据集适用于图像分类和图像分割任务,特别是在遥感、地球科学和洪水监测领域。
Kuro Siwo webdatasets is a global multi-temporal synthetic aperture radar (SAR) dataset for rapid flood mapping. It covers 43 flood events from 2015 to 2022 across six continents and three major climate zones worldwide. All flood area annotations are created by an expert team through meticulous image interpretation, with a spatial resolution of up to 10 meters. For each flood event, the dataset provides one post-disaster Sentinel-1 satellite image, two pre-disaster images, and corresponding digital elevation models (DEMs). The dataset includes SAR products at two processing levels: Ground Range Detected (GRD) and Single Look Complex (SLC), and comes with webdatasets required for the BlackBench benchmark, including training and test sets for both GRD and SLC products. The data is pre-divided into training and test sets, with the test set carefully selected to cover 10 flood events under diverse global environmental conditions to ensure challenging and representative evaluation. In terms of data structure, each sample in the GRD dataset includes VV and VH polarization bands for pre- and post-disaster images (totaling 6 band files), DEM, sample metadata, flood annotation masks (0: non-water, 1: permanent water, 2: flood), and valid pixel masks. Each sample in the SLC dataset includes pre- and post-disaster complex data images, DEM, metadata, and the same mask files. The dataset is suitable for image classification and segmentation tasks, particularly in remote sensing, earth sciences, and flood monitoring.
Kuro Siwo Webdataset 数据集详情
数据集基本信息
- 数据集名称: Kuro Siwo webdatasets
- 语言: 英语 (en)
- 任务类型: 图像分类 (image-classification)、图像分割 (image-segmentation)
- 标签: 地球科学 (earth-science)、遥感 (remote-sensing)、洪水制图 (flood-mapping)、SAR、山洪暴发 (flash-flood)、地球观测 (earth-observation)
- 许可证: CC BY 4.0 (cc-by-4.0)
数据集描述
Kuro Siwo 是一个全球多时相 SAR 数据集,用于快速洪水制图。数据集包含 2015-2022 年间发生在 6 大洲、3 个气候带的 43 个洪水事件。标注由专家团队通过精细的影像判读生成,空间分辨率为 10 米。每个洪水事件提供一幅 Sentinel-1 洪水后影像和两幅 Sentinel-1 洪水前影像,以及数字高程模型 (DEM)。为研究目的,数据集同时包含 GRD 和 SLC 两类产品。
- 资助方: 欧盟 Horizon Europe 研究与创新计划的 ThinkingEarth(授权协议号 101130544)和 MeDiTwin(授权协议号 101159723)
- 许可证: CC BY
数据集结构
样本字段
GRD webdatasets 每个样本包含以下字段
flood_vv.npy(float32): 洪水后影像的 VV 波段flood_vh.npy(float32): 洪水后影像的 VH 波段sec1_vv.npy(float32): 第一幅洪水前影像的 VV 波段sec1_vh.npy(float32): 第一幅洪水前影像的 VH 波段sec2_vv.npy(float32): 第二幅洪水前影像的 VV 波段sec2_vh.npy(float32): 第二幅洪水前影像的 VH 波段dem.npy(float32): 数字高程模型info.json: 样本元数据mask.npy(float32): 标签(0: 无水, 1: 永久水体, 2: 洪水)valid_mask.npy(float32): 有效像素掩膜(0: 无效, 1: 有效)
SLC webdatasets 每个样本包含以下字段
flood.npy(float32): 洪水后影像sec1.npy(float32): 第一幅洪水前影像sec2.npy(float32): 第二幅洪水前影像dem.npy(float32): 数字高程模型info.json: 样本元数据mask.npy(float32): 标签(0: 无水, 1: 永久水体, 2: 洪水)valid_mask.npy(float32): 有效像素掩膜(0: 无效, 1: 有效)
数据划分
数据集构建了一个具有挑战性的评估框架,选取全球 10 个洪水事件作为测试集,覆盖 Kuro Siwo 中所有六大洲和三个主要气候带的广泛环境条件。
- GRD 组件的训练集和测试集分别位于
train_GRD和test_GRD文件夹 - SLC 组件的训练集和测试集分别位于
train_SLC和test_SLC文件夹
引用信息
BibTeX 格式:
@inproceedings{NEURIPS2024_43612b06, author = {Bountos, Nikolaos Ioannis and Sdraka, Maria and Zavras, Angelos and Karavias, Andreas and Karasante, Ilektra and Herekakis, Themistocles and Thanasou, Angeliki and Michail, Dimitrios and Papoutsis, Ioannis}, booktitle = {Advances in Neural Information Processing Systems}, editor = {A. Globerson and L. Mackey and D. Belgrave and A. Fan and U. Paquet and J. Tomczak and C. Zhang}, pages = {38105--38121}, publisher = {Curran Associates, Inc.}, title = {Kuro Siwo: 33 billion m^{}2 under the water. A global multi-temporal satellite dataset for rapid flood mapping}, url = {https://proceedings.neurips.cc/paper_files/paper/2024/file/43612b0662cb6a4986edf859fd6ebafe-Paper-Datasets_and_Benchmarks_Track.pdf}, volume = {37}, year = {2024} }
APA 格式:
Bountos, N. I., Sdraka, M., Zavras, A., Karavias, A., Karasante, I., Herekakis, T., Thanasou, A., Michail, D. & Papoutsis, I. (2024). Kuro Siwo: 33 billion $ m^ 2$ under the water. A global multi-temporal satellite dataset for rapid flood mapping. Advances in Neural Information Processing Systems, 37, 38105-38121.
数据集卡片作者
Maria Sdraka




