基于高分遥感影像的地震受灾建筑物提取与倒损情况快速评估  

Damaged building extraction and rapid assessment for earthquake disasters based on high-resolution remote sensing images

作　　者：程希萌[1,2] 沈占锋[2] 邢廷炎[1] 董文[2] 

机构地区：[1]中国地质大学(北京)信息工程学院,北京100083 [2]中国科学院遥感与数字地球研究所遥感科学国家重点实验室,北京100101

出　　处：《自然灾害学报》2016年第3期22-31,共10页Journal of Natural Disasters

基　　金：国家高分辨率对地观测系统重大专项(03-Y30B06-9001-13/15-01);中科院重点部署项目(KZZD-EW-07-02);国家高技术研究发展项目(2013AA12A401)~~

摘　　要：以鲁甸地震为例,基于灾后无人机航拍影像,利用均值漂移分割算法以及C5.0决策树分类算法,提取了受灾建筑物区域,并对建筑物倒损率与地震烈度进行了估算,经目视解译验证以及与官方发布地震信息比对,倒损率相对误差均小于5%,地震烈度估计则与真实信息完全吻合。同时,利用采样方法计算了真实倒损面积与分类倒损面积的比值系数,提出了对于同一地震、相似地貌、相似建筑物分布的不同区域的建筑物倒损面积评估模型,经验证,其相对误差最小可低达5.31%,从而实现了倒损面积更准确的快速评估。In this study, disaster estimation for the Ludian region was conducted based on post-disaster UAV （un- manned aerial vehicle） aerophotographic images. By using the mean shift and the C5.0 decision tree algorithms, damaged buildings were extracted, the damaged building proportion was calculated, and the seismic intensity of study area was assessed. The verification results from visual interpretation and comparison with official data showed that, the relative error of damaged building proportion is less than 5 percent and the seismic intensity is the same as the official data. In addition, by calculating the ratio of the practical damaged building area to the classification re- suits of the damaged building area, a damaged building area evaluation model was proposed under the precondition that different areas are suffered from the same earthquake and have similar landform and building distribution. By verification, the minimum relative error of this model is 5.31 percent. The model proposed in this study can help to estimate damaged building area rapidly with high accuracy.

关 键 词：鲁甸地震 快速评估 建筑物提取 倒损率 倒损面积 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置] X43[自动化与计算机技术—控制科学与工程]