基于均值漂移和谱图分割的极化SAR影像分割方法及其评价  

Segmentation of PolSAR Data Based on Mean-Shift and Spectral Graph Partitioning and Its Evaluation

作　　者：赵磊[1,2] 陈尔学[1,2] 李增元[1,2] 冯琦[1,2] 李兰[1,2] 杨浩[1,2] 

机构地区：[1]中国林业科学研究院资源信息研究所,北京100091 [2]国家林业局遥感与信息技术重点开放性实验室,北京100091

出　　处：《武汉大学学报（信息科学版）》2015年第8期1061-1068,共8页Geomatics and Information Science of Wuhan University

基　　金：国家863计划资助项目(2011AA120404;2011AA120405)~~

摘　　要：提出了一种基于均值漂移和谱图分割的极化SAR(PolSAR)影像分割方法。首先,通过均值漂移算法对PolSAR影像进行过分割处理,并基于Wishart统计分布和假设检验的方法构建边缘检测器,充分利用了PolSAR影像的全极化信息提取边缘信息;然后,在过分割和边缘信息的基础上构建相似性度量矩阵,并采用归一化割准则实现PolSAR影像的分割。该算法充分利用了均值漂移算法过分割的特点,降低了谱图分割算法的运算代价,并结合谱图分割算法全局优化的优点改善了PolSAR影像的分割结果;最后,利用Radarsat-2全极化影像进行了实验,并采用改进的分割效果评价方法实现了精度评价。实验表明,该算法有效地实现了PolSAR影像的分割,显著提高了谱图分割算法的效率,分割结果优良,分割精度优于eCognition软件中的多尺度分割方法。A new segmentation approach of polarimetric synthetic aperture radar （PolSAR） data is pro- posed based on mean shift and spectral graph partitioning. First, Mean-shift algorithm is used to gen- erate the over-segmentation of PolSAR image. In order to extract edge information, we apply a set of detectors based on the Wishart distribution with a hypothesis testing method that has fully considered the polarization information in PolSAR images. Then, a similarity matrix is constructed based on the over-segmentation results and image edge information. The graph partitioning process is performed u- sing the normalized cut criterion. With this method, we improve the segmentation efficiency of spec- tral graph partitioning based on the over-segmentation results generated by Mean-shift. The quality of segmentation results is also improved as a result of the global optimization of spectral graph partitio- ning algorithm. We applied this method on Radarsat-2 full polarization images and evaluated the seg- mentation results. The experiement showed that this scheme can realize PolSAR segmentation effec- tively, speed up the original algorithm, and also demonstrates a better result than eCognition＇s Multi- resolution segmentation approach.

关 键 词：极化合成孔径雷达 影像分割 均值漂移 谱图分割 归一化割 

分 类 号：P237.3[天文地球—摄影测量与遥感] TP79[天文地球—测绘科学与技术]