HaitiBRD: A labeled satellite imagery dataset for building and road damage assessment of the 2010 Haiti earthquake

The dataset consists of paired pre- and post-disaster satellite imageries of the 2010 Haiti Earthquake,  a magnitude 7.0 earthquake that struck Haiti at 16:53 local time on 12 January 2010. The satellite imagery pairs were captured by GeoEye-1 satellite with high resolution (0.5m/pixel) and available from the Maxar Open Data program (https://creativecommons.org/licenses/by-nc/4.0/legalcode). The pre and post-disaster imagery were taken on 09 January 2010 and 15 January 2010, respectively.

The featured imageries (approximately 3.51 km2 in total) cover the vicinity of Port-au-Prince, the capital of Haiti, and the area situated about 25 kilometers west of the earthquake’s epicenter [18°28′N, 72°32′W]. These regions were severely affected by the earthquake, with the damaged seaport, roads, and buildings. In addition to pairs of pre- and post-disaster satellite images of the event, this dataset provides the label of the damage to buildings and roads. Over 9000 building segments are classified into four damage levels (“No damage,” “Minor,” “Major,” and “Destroyed”) and illustrated in different colors in the segmentation masks (.png). On the other hand, the undamaged and damaged road segments are denoted by line strings in .shp files. All annotations were done by the in-house team with quality control.

本数据集包含2010年海地地震的配对式灾前与灾后卫星影像。此次地震震级为7.0，于2010年1月12日当地时间16:53袭击海地。该影像对由GeoEye-1卫星拍摄，分辨率达0.5米/像素，可通过Maxar开放数据计划获取（https://creativecommons.org/licenses/by-nc/4.0/legalcode）。灾前影像拍摄于2010年1月9日，灾后影像拍摄于2010年1月15日。

本次重点展示的影像总覆盖面积约3.51平方千米，涵盖海地首都太子港周边区域，以及距离地震震中[北纬18°28′、西经72°32′]以西约25千米的区域。该区域受此次地震影响严重，港口、道路与建筑均遭损毁。除本次地震的灾前、灾后卫星影像配对数据外，本数据集还提供建筑与道路损毁标注。超过9000个建筑地块被划分为四个损毁等级：“无损坏”“轻微损坏”“严重损坏”与“完全损毁”，并在分割掩码（segmentation masks，.png格式）中以不同颜色标注。另一方面，未损毁与损毁的道路地块则以.shp格式文件中的线要素表示。所有标注均由内部团队完成并经过质量管控。

搜集汇总

背景与挑战

背景概述

HaitiBRD数据集包含2010年海地地震前后的高分辨率卫星图像对，覆盖了太子港附近约3.51平方公里的区域。数据集提供了超过9000个建筑物段和道路段的损坏标签，建筑物损坏分为四个等级，道路损坏分为未损坏和损坏两类。所有标注均由内部团队完成并经过质量控制。