用于高分辨率SAR影像建筑物提取的对象级高亮特征描述方法

目的 合成孔径雷达(synthetic aperture radar,SAR)影像分辨率的不断提高为建筑物提取提供了有效的数据支持,而传统像素级方法,提取建筑物的效果较差,精度较低。通过分形网络演化分割算法(fractal net evo-lution approach,FNEA)获取分析单元,利用对象级分析单元与邻近环境之间的上下文特征,提出了高亮邻接强度特征(highlight adjacent intensity,HAI)与亮点散射密度特征(shining point distribute density,SDD)的概念,然后结合上述两种特征进行对象级建筑物的提取。最后通过几组实验验证了基于面向对象特征方法比面向像素特征方法对高分辨率SAR建筑物提取具有更好的效果、更高的精度。

Object-Level Highlight Features For High-resolutionSAR Building Extraction

Objective With the improvement of SAR image resolution,it is now increasingly used as an effectivedata support for building extraction.However,the traditional pixel-based method is not practicable inbuilding extraction.Not only are the result not amiable,but also the accuracy is very poor.So in thispaper we will firstly apply FNEA algorithm(fractal net evolution approach)on SAR image to obtainanalysis units.Then contextual feature of those object-level units will be utilized to propose the con-ception of highlight adjacent intensity(HAL)and shining point distribute density(SDD).After thatthese two features will be combined to be used in the process of object-level building extraction.Final-ly,a couple of experiments are conducted show that an object-oriented method outperforms pixel-based methods in building extraction from high-resolution SAR images.