基于广义高斯模型多时相SAR图像非监督变化检测方法.pdfVIP

基于广义高斯模型的多时相SAR图像非监督变化检测方法 何敏 何秀凤 河海大学卫星及空间信息应用研究所，南京，210098 摘要：提出了一种基于广义高斯模型的多时相SAR 图像非监督变化检测方法。首先估计变 化和未变化像元的概率密度参数，然后采用改进的KI （Kittler–Illingworth ）算法计算最佳分 割阈值并提取变化区域，最后以南京地区的ERS-2 SAR 影像为例进行了试验研究。试验结 果表明，所提出的变化检测方法比简单高斯模型更准确地描述差异影像像元的分布，并且变 化阈值的选取更方便。 关键词：广义高斯分布（GGD ）；变化检测；KI 阈值选取；多时相SAR 图像 中图分类号：TN957.52 Unsupervised Change Detection in Multitemporal SAR Images based on the Generalized Gaussian Model Min He Xiufeng He Institute of Satellite Navigation Spatial Information System, Hohai University, Nanjing, 210098 Abstract: A novel unsupervised change detection approach oriented to the analysis of multitemporal SAR images is presented. This approach is based on the Generalized Gaussian distribution Model. Under the assumption that the changed pixels and unchanged pixels follow Generalized Gaussian distribution (GGD), the probability density of the two classes pixels are estimated. The improved Kittler–Illingworth (KI) threshold selection criterion is derived under the generalized Gaussian assumption for modeling the distributions of changed and unchanged classes. The optimal threshold is calculated by the improved KI threshold selection criterion. Then the changed map is extracted based on the optimal threshold. The tests have been conducted by using ERS-2 SAR images from Nanjing city. The results show that the approach proposed has better performances than simple Gauss model in modeling the distributions of difference image and benefits changed threshold selection. Keywords: Generalized Gaussian