一种新的极化SAR图像非监督分类算法研究

基金项目: 国家自然科学基金资助项目(40801183,60890074)；国家863计划资助项目(2007AA12Z180,2007AA12Z155)

详细信息

An Unsupervised Classification Method of POLSAR Image

¹ School of Electronic Information,Wuhan University,129 Luoyu Road,Wuhan 430079,China

Funds: 国家自然科学基金资助项目(40801183,60890074)；国家863计划资助项目(2007AA12Z180,2007AA12Z155)

摘要: 提出了一种建立在Mean-shift过分割结果区域图上的极化SAR图像非监督分类算法。首先通过Mean-shift算法得到极化SAR图像的过分割结果区域图,并将过分割小块视为“超级像素”,然后在Freeman-Durden分解的基础上引入散射功率熵和各向异性量参数来进一步分析“超级像素”的混合散射机制问题,最后结合Wishart迭代聚类实现极化SAR图像的非监督分类。实验表明,该算法具有较为满意的分类效果。
- Mean-Shift /
- 非监督分类 /
- Freeman-Durden分解
Abstract: An unsupervised classification algorithm established on the mean-shift over-segmentation is presented in this paper. First,an over-segmentation result is obtained by a mean-shift algorithm and the segmentation patches are treated as "super-pixels". Then,based on Freeman-Durden decomposition,we survey the four different combinations of three basic scattering mechanisms by introducing two new parameters-the scattering power entropy and anisotropy. Finally,the iterative wishart classifier is applied to get the final classification results. The effectiveness of this algorithm is demonstrated by using the German aerospace center's (DLR) E-SAR L-band polarimetric synthetic aperture radar images.
- mean-shift /
- unsupervised classification /
- Freeman-Durden decomposition