高分遥感影像震害提取的变化检测算法改进

An Improved Change Detection Algorithm for Building Damage Extraction Based on High Resolution RS images

本文针对遥感多时相变化检测应用于地震灾害信息提取中存在的问题,提出一种改进算法,通过搜索震后影像上建筑物在震前影像上对应区域周边一定范围内的影像相关性最大值(区域寻优)来代替传统上严格对应像元的相关系数值,以减少因配准精度、摄影方位、地表高程、建筑物高度等造成的同名地物变形对变化检测结果的影响,提高基于相关系数变化检测的精度。以都江堰市城区在2008年汶川地震震前的QuickBird影像和震后的航空遥感影像为实验对象,采用本文方法进行了震害变化检测实验,并与传统方法进行了比较,表明改进的算法能够提高震害信息提取的精度且具有一定的稳定性。

In this paper, an improved method based on digital change detection of multitemporal high resolution RS images is proposed to extract the building damage caused by catastrophic earthquakes, which aims to improve the accuracy of classification and to reduce the impact of the deformation of corresponding image point caused by low registration precision, different photographic mode, and the differences of surface elevation and height of buildings by searching the maximum image correlation coeffieient of a point on post-event image from a certain spatial domain around the registration point on the pre-event image instead of the conventional correlation coefficient exactly on the the registration point. The improved method is applied to detect the changes of buildings in a test area of the city of Dujiangyan after the 2008 Wenchuan earthquake by using pre-earthquake QuickBird remote sensing data acquired on July 22, 2005 and post-earthquake aerial images acquired on May 18, 2008. The results of the experiments are compared with the results using traditional method and indicate that the improved method is more precise and stable to extract the seismic damage information than traditional method.