基于能量最小化的星载SAR图像建筑物分割方法  

Building Segmentation of Spaceborne SAR Images Based on Energy Minimization

作　　者：周则明[1] 孟勇[1] 黄思训[1] 胡宝鹏 

机构地区：[1]解放军理工大学气象海洋学院,南京211101

出　　处：《自动化学报》2016年第2期279-289,共11页Acta Automatica Sinica

基　　金：国家自然科学基金(61473310;41174164;41175025;41275029;41305138);公益性行业(气象)科研专项(GYHY201306068)资助~~

摘　　要：针对星载合成孔径雷达(Synthetic aperture radar,SAR)图像信噪比低、建筑物目标几何变形大以及周围背景复杂的特点,本文提出了一种基于能量最小化的星载SAR图像建筑物分割方法.基于星载SAR图像数据构造条件概率能量项,推动变形曲线向建筑物目标边界演化;在能量泛函模型中定义长度能量项以保证变形曲线的平滑;在水平集方法获取的SAR图像初始分割结果的基础上,以高分辨率光学遥感影像中建筑物目标的轮廓作为先验信息,构造先验形状能量项约束曲线在第二阶段的演化,最终实现SAR图像建筑物的分割.实验结果表明,该方法显著提高了建筑物目标轮廓的分割精度.To aim at the difficulties of low signal to noise ratio（SNR） of spaceborne synthetic aperture radar（SAR） image,high geometric deformation and the complex background of buildings, a variational method is proposed for segmenting the building of spaceborne SAR image based on energy minimization. The condition probability energy term is defined with the data of SAR image to drive the deformation curve to the boundary of the building. The length energy term is constructed with the length of the evolving contour to ensure the smoothness of the deformation curve. The prior shape of the building is acquired from the corresponding high-resolution optical remote sensing image and the prior shape energy term is introduced into the energy functional. On the basis of a preliminary segmentation result which is achieved by the level set method, the building segmentation is implemented eventually by imposing the constraint of the prior shape.Experimental results demonstrate that the proposed model can improve the segmentation accuracy of the building target significantly.

关 键 词：星载SAR图像 建筑物分割 能量最小化 先验形状 

分 类 号：TN957.52[电子电信—信号与信息处理]