基于协同分割的高分辨率遥感图像变化检测  

Change detection from high-resolution remotely sensed images based on cosegmentation

作　　者：袁敏[1,2,3] 肖鹏峰[1,2,3] 冯学智[1,2,3] 张学良[1,2,3] 胡永月 

机构地区：[1]江苏省地理信息技术重点实验室,南京大学,南京210023 [2]卫星测绘技术与应用国家测绘地理信息局重点实验室,南京大学,南京210023 [3]南京大学地理信息科学系,南京210023

出　　处：《南京大学学报（自然科学版）》2015年第5期1039-1048,共10页Journal of Nanjing University（Natural Science）

基　　金：江苏高校“青蓝工程”(201423);浙江省科技计划(2014F50022)

摘　　要：针对面向对象变化检测存在的多时相对象边界不一致的难题,基于计算机视觉领域的多视图像协同分割思想,提出一种面向多时相高分辨率遥感图像变化检测的协同分割方法.首先对多时相遥感图像进行协同处理,利用多时相信息发现变化特征,以光谱变化为指标,获得变化强度图,进而在变化强度图的引导下,结合各时相的图像自身特征进行分割,通过能量函数的构建和优化,直接生成边界准确、空间对应的多时相变化对象.利用两个时相的高分辨率航空图像进行实验表明,该方法可以较完整准确地分割出变化对象,通过建立其空间对应关系,能清晰地表达对象的变化过程,为高分辨率遥感图像面向对象变化检测提供了新思路.Due to the inconsistency of multi-temporal objects＇ boundaries for object-based change detection,this paper proposes a new change detection approach from multi-temporal high-resolution remotely sensed images based on the concept of cosegmentation in the filed of computer vision.First,multi-temporal remotely sensed images are co-processed to discover the change feature and a map of change intensity is obtained using the magnitude of spectral change between images.Then cosegmentation is performed under the guidance of the change intensity map,combined with each image features.Multi-temporal change objects with accurate boundaries and spatial correspondence are directly generated by energy function minimization finally.Experimental results obtained on multi-tem-poral aerial images show that multi-temporal change objects are preferably segmented and the change process can also be clearly acquired through establishing a correspondence between multi-temporal change objects.This novel method can provide a workable way for object-based change detection from high-resolution remotely sensed images.

关 键 词：高分辨率遥感图像 多时相图像 变化检测 协同分割 最小割/最大流 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]