一种基于概率图模型的多时相SAR相干变化检测方法  

A Multi-temporal SAR Coherent Change Detection Method Based on Probabilistic Graphical Models

作　　者：冀广宇 董勇伟[1] 李焱磊[1] 梁兴东[1] 

机构地区：[1]中国科学院电子学研究所微波成像技术重点实验室,北京100190 [2]中国科学院大学,北京100049

出　　处：《电子与信息学报》2017年第12期2912-2920,共9页Journal of Electronics & Information Technology

摘　　要：相干变化检测(CCD)利用重轨SAR数据对场景中表现为低相干特性的变化区域具有良好的检测性能,然而场景中诸如植被、阴影、强散射旁瓣、低散射等区域也呈现低相干特性,对检测结果造成干扰,尤其在高波段SAR CCD中,对检测效果影响更加明显。该文利用多时相SAR数据形成的相干变化差异图像(CCD图像)建立概率图模型,提出一种多时相CCD处理方法。该方法以多时相CCD图像作为观测量,通过选取合适的参与处理图像数量及优化场景中变化区域的分类,计算目标变化区域的后验概率,可有效减小低相干干扰区域造成的影响。仿真和实测数据结果验证了该方法的正确性和有效性。Coherent Change Detection （CCD） has good performance on detecting change regions with low coherence in the scene by using repeat-pass Synthetic Aperture Radar （SAR） data. However, some regions as vegetation, radar shadows, sidelobes of strong reflectivity and low reflectivity areas show low coherent character as well, which disturbs the result of change detection, especially in high frequency band SAR CCD with more evident disturbance. This paper proposes a multi-temporal CCD method by establishing a probabilistic graphical model using CCD images formed by multi-temporal SAR data. In this method, multi-temporal CCD images are used as observations to calculate a posterior probability of objective change region via choosing appropriate number of processing CCD images and optimizing the classification of change regions in the scene. The proposed method can reduce the disturbance of low coherence disturb regions effectively. The simulated and experimental results demonstrate the validity and effectiveness of the proposed method.

关 键 词：重轨合成孔径雷达 相干变化检测 多时相 概率图模型 后验概率 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]