一种基于证据理论的多极化SAR图像分割方法

Segmentation of multi-polarized SAR imagery based on the theory of evidence

本文提出了一种基于证据理论的融合像素信息和上下文信息的多极化合成孔径雷达(SAR)图像分割方法。D-S证据理论是一种不确定性推理方法。基于D-S的SAR影像分割方法将像素信息和上下文信息看作两类证据,先对高度平滑后的影像作初始的过分割,然后基于D-S理论对初始分割图斑的边界进行迭代修正,最后再融合两类证据对初始分割的图斑进行合并。本文方法比传统基于像素信息的方法具有更强的抗噪性能。横断山脉的双极化SAR影像分割实验表明,该方法对噪声敏感度相对较低,可以获得较可靠的分割结果。

An algorithm based on the theory of evidence fusing pixel and contextual information for multi-polarized SAR images was proposed in this article. D-S method is a kind of indefinite inference theory. Pixel information and contextual information were considered as two types of evidence in the segmentation for SAR imagery based on D-S theory. Firstly the highly smoothed image was oversegmented ; secondly the borderlines of each segment were modified by fusing two kind of evidence-pixels＇ values and information of their neighbor areas ; finally initial segments were merged based on the fusion of two evidences. The information of each pixel was considered as a kind of evidence, and information of pixels＇ neighbors was considered as a new kind of evidence. Therefore it could perform better than traditional algorithm based on pixels＇ value when images with rich noise are processed. The experimental results on dual-polarimetric SAR images in Hengduan Mountain showed that the algorithm was valid with less sensitivity to speckle noise.