基于双边加权核图割的SAR图像变化检测  

SAR Image Change Detection Based on Bilateral Weighted Kernel Graph Cuts

作　　者：丁稀 董张玉[1] 杨学志 DING Xi;DONG Zhang-yu;YANG Xue-zhi(School of Computer Science and Information Engineering,Hefei University of Technology/Anhui Province Key Laboratory of Industry Safety and Emergency Technology,Hefei 230601;School of Software,Hefei University of Technology/Intelligent Interconnected Systems Laboratory of Anhui Province,Hefei 230601,China)

机构地区：[1]合肥工业大学计算机与信息学院/工业安全与应急技术安徽省重点实验室,安徽合肥230601 [2]合肥工业大学软件学院/智能互联网系统安徽省实验室,安徽合肥230601

出　　处：《地理与地理信息科学》2022年第3期37-42,共6页Geography and Geo-Information Science

基　　金：安徽省重点研究与开发计划项目(202004a07020030);中央高校基本科研业务费专项项目(JZ2021HGTB0111)。

摘　　要：为解决SAR图像因相干斑噪声严重而影响变化检测结果的问题,该文从像素级和区域级两个角度对SAR图像的空间邻域关系进行研究,提出一种基于双边加权核图割的SAR图像变化检测算法。首先通过均值比算子生成差异图,并利用核函数隐式地将差异图转换为图割能量函数的一元项;然后分别针对邻域内像素的空间上下文关系与以像素为中心的区域空间上下文关系进行研究,将其量化为图割能量函数的二元项。选取Ottawa和Mexico fire两组SAR数据集对算法进行验证,该算法平均总体精度为98.44%,Kappa系数相比PCA-Kmeans、FLICM、MRFFCM、KGC算法分别提高了4.64%、6.16%、4.15%和5.04%,证明该算法能够提高变化区域和非变化区域之间的可分离性,对噪声严重的SAR图像的抗噪能力较强,且图像边缘定位准确。In order to solve the problem of serious speckle noise in synthetic aperture radar(SAR)image,this paper studies the spatial neighborhood relationship of SAR image from pixel level and region level,and proposes a change detection algorithm based on bilateral weighted kernel graph cuts.The algorithm first generates difference graph by mean ratio operator.Secondly,the kernel function is used to transform the difference graph into unary term of the energy function of graph cuts implicitly,and the spatial context relations of the pixels in the neighborhood and the regions centered on the pixel are studied respectively,which are quantified as the binary term of the energy function of graph cuts.Experimental results show that the overall accuracy of this algorithm is as high as 98.44%,and compared with traditional PCA-Kmeans,FLICM,MRFFCM and KGC algorithms,the Kappa coefficient is improved by 4.64%,6.16%,4.15%and 5.04%respectively.The proposed algorithm can improve the separability between the changing region and the non-changing region,and has strong anti-noise ability for SAR images with serious noise and accurate edge location.

关 键 词：核图割 SAR图像变化检测 上下文信息 边缘相似度 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]