Labeled Dataset of SAR Sentinel-1 Imagery Despeckled with Multitemporal Fusion (5–25 Looks, VV/VH Polarizations)

This dataset contains globally distributed Sentinel-1 SAR image patches (512×512 pixels) acquired using dual polarizations (VV and VH) and both ascending and descending orbits.

Each sample includes:
(i) Single-acquisition SAR images (noisy input)
(ii) Ground-truth images generated through multitemporal averaging using 5, 10, 15, 20, and 25 acquisitions
(iii) Corresponding ESA WorldCover land cover maps
(iv) A metadata CSV file containing geographic coordinates and Google Earth Engine identifiers.

The dataset is designed for supervised training and benchmarking of SAR despeckling algorithms and multitemporal fusion strategies.

本数据集包含全球分布的Sentinel-1合成孔径雷达（Synthetic Aperture Radar，SAR）图像块，尺寸为512×512像素，采用双极化（VV与VH）模式采集，且覆盖升轨与降轨两种轨道方向。

每个样本包含以下内容：
(i) 单次采集SAR图像（带噪输入样本）
(ii) 利用5、10、15、20、25次多时间序列采集结果进行平均生成的真值图像
(iii) 对应欧洲空间局（European Space Agency，ESA）WorldCover土地覆盖图
(iv) 包含地理坐标与谷歌地球引擎（Google Earth Engine）标识符的元数据CSV文件。

本数据集旨在为SAR去噪算法与多时间融合策略的监督训练及性能基准测试提供支撑。