在机载高分辨率X波段雷达影像上,城市水体、水泥道路、裸露土壤等地物回波信号弱,具有较强的相似性.目前的H/Alpha-Wishart 、Freeman-Durden 等系列算法对这些地物的区分不理想.本文针对X波段的全极化数据,提出基于预分类再分割的方法,重新估计极化协方差矩阵;然后根据协方差矩阵的熵值与同极化通道相位差的标准差构建特征空间,利用阈值法方法进行特征空间分割,获取最终精细分类结果.实验证明,本文方法可以在高分辨率条件下对水体、水泥道路、裸露土壤进行较为精确的划分,总体分类精度优于80% 、Kappa 系数高于0.7 ,是一种有效的低后向散射地物精细分类方法.

In high resolution X-band POLSAR image, the water body, cement roads and the bare soil are always at low level radar backscattering signals, which is caused by no Bragg scattering phenomenon in smooth surfaces. The prevalent H/Alpha- Wishart and Freeman-Durden methods cannot distinguish those targets. This paper explores the improved X-band classif ication algorithm based on the pre-classif ication result for low backscattering objects in urban areas. The occurrence plane which is combined by entropy and the standard deviation of the co-pol channel diff phase is used to ref ine the pre-classif ication. The experiments show that the overall accuracy is above 80% and Kappa coeff icient is higher than 0.7. The improved method improves the potential to distinguish the mixture classes of the low backscattering objects.