基于灾后无人机遥感影像的地震损毁建筑物样本集构建
Construction of a Sample Set of Earthquake-Damaged Buildings Based on Post-Disaster UAV Remote Sensing Images

灾后建筑物损毁评估可以高效辅助应急救援、指挥决策与恢复重建等工作。近年来，遥感技术与深度学习方法的飞速发展为高效掌握灾后建筑物的损毁情况提供了重要的技术支撑。然而，高分辨率遥感影像上的损毁建筑物样本获取困难，且可用于深度学习模型训练的公开数据集较少。因此，本文利用2021年云南大理漾濞6.4级地震灾前Google Earth 20级影像和灾后高分辨率无人机遥感影像分别进行建筑物轮廓自动提取，通过对比分析灾前、灾后建筑物分布情况与灾后建筑物的屋顶损毁面积占比，并结合灾后三维模型辅助研判，最终得到一组带有损毁等级属性的灾后损毁建筑物样本数据集，为基于深度学习方法进行建筑物损毁评估的模型训练等算法研究工作提供数据基础。
Assessment of damaged buildings in post-disaster images can efficiently assist emergency rescue, command decision-making, restoration and reconstruction. In recent years, with the rapid development of remote sensing technology and deep learning methods, there has been important technical support for efficiently grasping the damage of buildings after disasters. However, samples of damaged buildings on high-resolution remote sensing images are difficult to obtain, and there are few public datasets available for deep learning model training. Therefore, this paper uses the Google Earth 20 image before the 2021 Yangbi M6.4 earthquake in Dali, Yunnan Province and the post-disaster high-resolution UAV remote sensing image to extract the building contour automatically respectively. By comparing and analyzing the distribution of buildings before and after the disaster and the proportion of roof damage area of buildings after the disaster, and combining with the post-disaster three-dimensional model to assist in research and judgment, a set of sample data sets of post-disaster damaged buildings with damage grade attribute are finally obtained. It provides data basis for algorithm research such as model training of building damage assessment based on deep learning method.