基于DInSAR技术监测地震形变及深度学习提取损毁建筑物——以厄瓜多尔为例  

Monitoring Earthquake Deformation Based on DInSAR Technology and Extracting Damaged Buildings by Deep Learning--Take Elguado as An Example

作　　者：郭天豪 解斐斐[1,2] 霍志玲 GUO Tianhao;XIE Feifei;HUO Zhiling(College of Geomatics, Shandong University of Science and Technology, Qingdao Shandong 266590, China;Shandong 3S Engineering Technology Center, Qingdao Shandong 266590, China)

机构地区：[1]山东科技大学测绘科学与工程学院,山东青岛266590 [2]山东省3S工程技术中心,山东青岛266590

出　　处：《北京测绘》2021年第9期1225-1229,共5页Beijing Surveying and Mapping

基　　金：2019年国家级大学生创新创业训练计划项目(201910424033)。

摘　　要：针对地震导致的地表变形、建筑物损坏等问题,利用合成孔径雷达差分干涉测量技术检测受灾区,并利用深度卷积网络提取损坏建筑物。以2016年4月16日厄瓜多尔地震为例,利用Sentinel-1A雷达数据获取地震形变。结果表明:厄瓜多尔地区整体地震形变在0~0.226 m,可检测形变变化为0.028 m;结合光学遥感影像进行分析,结果表明:80%损坏严重的建筑物存在于形变变化区域,形变较大区域道路受阻更为严重。基于高分辨率Worldview2光学卫星数据,利用ENVINet5深度卷积网络提取损坏建筑物,分类精度为74%,验证其在震后建筑物提取应用的可行性。Aiming at the problems of ground deformation and building damage caused by earthquake,thepaper applied DInSAR to detect the affected areas and ENVINet5 to extract damaged buildings.It took the earthquake in Ecuador on April 16,2016 as an example and obtained the seismic deformation by using Sentinel-1Aradar data.As a result,the overall seismic deformation in Ecuador was between 0 and 0.226 m,and the detectable deformation change was 0.028 m.Combinedwith optical remote sensing image:we could know the probability of seriously damaged buildings exists in the deformation region was 80%,and the road obstruction was more serious than buildings.Based on Worldview2,the high resolution optical satellite data and using ENVINet5,the classification accuracy of damaged buildings was 74%,which verified the feasibility of its application in building extraction after earthquake.

关 键 词：合成孔径雷达干涉测量 地震形变 深度学习网络 建筑物信息提取 

分 类 号：P258[天文地球—测绘科学与技术]