基于高斯混合模型的SAR图像变化检测方法  

SAR Image Change Detection Based on Gaussian Mixture Model

作　　者：王霞[1] 刘振华[1] 于俊朋[1] 邓楚强[1] 

机构地区：[1]南京电子技术研究所,南京210039

出　　处：《现代雷达》2014年第9期34-35,78,共3页Modern Radar

摘　　要：通过对数比值法构造合成孔径雷达(SAR)图像的差异图,然后分析差异图的概率统计,提出了一种基于高斯混合模型的SAR图像变化检测方法。研究分析表明,采用四个高斯函数的混合模型(方差较大的两个代表变化类,另外两个代表未变化类)可以较好地拟合差异图的概率统计分布。文中提出的变化检测方法,实现了非监督检测,减少了运算代价。对实测数据的处理结果显示,该方法可以得到好的变化检测结果。The difference image was generated by the log-ratio technique. A SAR image change detection approach based on Gaussian mixture model was proposed by analysis of the statistical distributions about the difference image. The mixture model by employing four Gaussian distributions （the two distributions with larger variance were defined as the changed, and the others were defined as the unchanged） was suitable fitting. Experimental results showed that the changed targets between SAR images were successfully and unsupervisedly detected by the proposed approach.

关 键 词：变化检测 合成孔径雷达图像 高斯混合模型 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]