高分辨率极化SAR图像水平集分割

目的 传统的极化SAR图像分割方法中,由于采用的统计分布模型不能较好地描述高分辨率的图像纹理特征,导致高分辨率极化SAR图像分割效果较差。针对这个问题,本文将具有广泛适用性的KummerU分布嵌入到水平集极化SAR图像分割方法中,提出了一种新的极化SAR图像分割算法。方法 将KummerU分布作为高分辨率极化SAR图像的统计模型,定义一种适用于极化SAR图像分割的能量泛函;利用最大似然法对各个区域的KummerU分布进行参数估计,并通过数值偏微分方程的方法求解水平集函数,实现极化SAR图像的区域分割。结果 分别对仿真全极化数据,真实全极化数据进行分割实验,结果表明本文提出的方法其分割精度高于传统方法,分割精度高于95%,从而验证了新方法的有效性。结论 本文算法能够对各向同质区和各向异质区的极化SAR图像都能取得良好的分割效果,并适应于多种场景,有效地分割出背景和目标。

High-resolution PolSAR image level set segmentation

Objective Statistical models cannot agree well with the distribution of texture in a high-resolution synthetic aperture radar (SAR) image. Obtaining good results with traditional polarimetric SAR (PolSAR) image segmentation methods is therefore difficult. To overcome this problem, we propose a new PolSAR image segmentation method that combines KummerU distribution and the level set framework. Method The proposed method defines a new energy function with the KummerU probability density function as a high-resolution PolSAR image statistical model. The method is therefore suitable to PolSAR image segmentation. To implement PolSAR image segmentation, the parameters of KummerU distribution are estimated with the maximum likelihood method. The level set function is applied to the numerical solution of a partial differential equation. Result Experiments are based on synthetic-full-polarization and real-full-polarization SAR images. All experiments show good results, with an accuracy level above 95% compared with that of the traditional method and which therefore proves the applicability of the proposed algorithm. Conclusion We propose a novel PolSAR image segmentation method based on a level set framework. The algorithm is applicable to high-resolution PolSAR images. Experiment results indicate that our model can be effectively used in most scenes and can separate targets from the background in homogeneous, heterogeneous, and extremely heterogeneous areas.