x-harvey

This dataset captures the impact of Hurricane Harvey, a Category 4 hurricane that caused catastrophic flooding in Texas and Louisiana in August 2017, resulting in over 100 fatalities. It focuses on affected areas of Houston, Texas, originally divided into east and west regions with 469 and 461 Sentinel-2 image tiles, respectively. Each tile includes pre- and post-flood Sentinel-2 images, SAR coherence and intensity images, and building footprints marking damaged and undamaged structures before and after the disaster, provided at three spatial resolutions (1m, 2m, and 10m). Building footprints were sourced from OpenStreetMap, and flooded building labels were crowd-sourced via the MAXAR open data program. In this version, the original train/test split has been adjusted to new train, validation, and test sets to reduce distribution shifts and improve model evaluation. The very high-resolution WorldView imagery (2262×1972 pixels at 1m resolution) has been excluded. This dataset is used primarily for segmentation and change detection tasks related to flood impact assessment.

本数据集记录了2017年8月于美国得克萨斯州与路易斯安那州引发灾难性洪涝的四级飓风哈维（Hurricane Harvey）所造成的影响，此次灾害造成超100人遇难。数据集聚焦得克萨斯州休斯顿的受影响区域，该区域最初被划分为东西两片子区域，分别包含469张与461张哨兵二号（Sentinel-2）影像瓦片。每张影像瓦片均涵盖灾前与灾后的哨兵二号影像、合成孔径雷达（Synthetic Aperture Radar，SAR）相干性影像与强度影像，以及标注灾害前后受损与未受损建筑的建筑轮廓（building footprints），相关数据提供了1米、2米与10米三种空间分辨率。建筑轮廓数据源自开放街道地图（OpenStreetMap），洪涝受损建筑的标注信息则通过MAXAR开放数据项目以众包方式采集。本版本对原始的训练集/测试集划分方式进行了调整，重新划分为训练集、验证集与测试集，以降低数据分布偏移，提升模型评估效果。分辨率为1米、尺寸为2262×1972像素的超高分辨率WorldView影像已被剔除出本数据集。本数据集主要应用于与洪涝影响评估相关的分割与变化检测任务。