全极化SAR遥感图像中相干斑滤波机理及方法研究

Speckle filtering is always the difficulty of fully polarimetric SAR imagery processing. Most classic methods, which have not taken into full account the spatial and polarimetric properties, can only enhance the one aspect of performance. Thus the accuracy of decomposition and classification yield by these methods will decrease heavily. The primary mission of this proposal can be divided into two aspects. Firstly, the mechanism for speckle filtering will be examined firstly in this proposal. Then, the polarization fork will be used to discriminate point targets pixels. After a thorough survey about the current methods, a revised four-component decomposition method will be developed, which will be used to classify the homogeneous pixels into several category. Secondly, the nonlinear anisotropic diffusion model will be improved for the filtering of PolSAR imagery. This filtering procedure is an iterative processing schema, which objects are the homogeneous pixels exported in the last step. By the aid of this research program, the accuracy of decomposition and classification applied to fully polarimetric SAR imagery will increase. The research results will be beneficial to the analysis and application of remote-sensing images by airborne and spaceborne PolSAR system.

相干斑滤波一直是全极化SAR遥感图像预处理的难点。传统方法大多仅能提高某一方面的滤波性能，并未统筹考虑空域细节信息和极化属性的保持，这将严重降低后期极化分解、分类、目标识别等处理算法的精度。本课题的主要工作包括：（1）通过分析相干斑滤波的内在机理，基于极化叉属性完成确定性的点散射目标的分离，提出对四分量模型的改进方案，并将其应用于分布式散射目标像素的初分类；（2）改进了非线性各向异性扩散滤波模型，在经过初分类的同质像素集合上完成迭代式的相干斑滤波。本课题的研究结果能够极大的提高全极化SAR遥感图像后期处理算法的精度，对于机载和星载平台下全极化SAR遥感图像的分析及应用具有重要的理论意义和实用价值。

项目对全极化合成孔径雷达图像的相干斑的内在机理进行了研究，结合几何扰动滤波检测点散射像素的基本原理，实现了一种改进的点散射目标的检测方法；同时，项目组经过调研和研讨，采用了TOUZI分解方法完成同质像素分类；然后，设计并实现了张量扩散的基本模型和规范化模型，项目组通过申请获得了超高分辨率机载平台的全极化数据，开展了一系列的定性和定量的实验，验证了项目组提出的各向异性扩散滤波算法在等效视数的提高上具有较好性能；同时，滤波之后的全极化数据仍然能够保持原始数据的极化属性和图像的细节。本项目研究成果可以推广到实际应用，例如：海岸带监测。