场景自适应的高分辨率全极化雷达影像去噪方法

Polarimetric SAR (PolSAR) systems have their irreplaceable merits for earth observation. However, due to the limitation of imaging mechanism, PolSAR images are inherently affected by speckle noise, which is more serious and complicated in high-resolution systems, degrading the quality of images and restricts their potentials in many applications to a large degree. This project is to develop a scene adaptive filtering method by employing the traits of the speckle in high-resolution data: a heterogeneity index is investigated to assess the complexity of textures in image, and the scenes of the image blocks are then categorized into homogeneous areas, textural areas and the transitional areas; the traits of the fully-developed and nonfully-developed speckle are respectively utilized for the similar block matching of corresponding areas; by employing the patch-based collaborative processing framework, the adaptive filtering method is developed for the similar block groups with different type of scenes, aiming at suppressing the speckle to a large degree and effectively retaining the image details. This study has the possibility to provide a new processing framework for high-resolution PolSAR image despeckling, improving the quality of images and upgrading the potential of PolSAR data in many applications. Therefore, this research has important theoretical and applied significances.

全极化雷达系统具有不可替代的对地观测优势，但由于成像机理的限制，使影像不可避免存在相干斑噪声，特别是在高分辨率极化系统中更加突出和复杂，严重降低了影像质量，制约了其在各方面的应用潜力。本项研究针对高分辨率数据相干斑的独特性质，发展场景自适应的滤波方法：研究衡量影像场景纹理复杂性的异质性指数，实现影像同质区、纹理区及过渡区域的自动确定；利用完全发育和非完全发育噪声的特性，分别实现相应区域相似影像块的匹配聚合；借鉴图块协同计算框架，发展针对不同场景类型相似块组的自适应滤波方法，在显著抑制噪声的同时，有效保持影像细节信息。本研究可为高分辨率全极化遥感影像去噪提供新的研究思路与处理方法，提升影像质量，增强极化雷达数据在各领域中的应用潜力，具有重要的理论与应用价值。

全极化雷达系统具有独特的对地观测优势，但受限于固有成像机理，所获影像不可避免存在相干斑噪声，导致影像质量降低。高分辨率极化噪声问题更加显著和复杂，制约了其在各方面的应用潜力。本项研究利用高分辨率数据相干斑的独特性质，借鉴图块匹配处理框架，发展场景自适应的块匹配滤波方法。研究团队在基金的资助下，发展了两类块匹配滤波方法。首先，提出了一种结合相似块匹配策略及线性最小均方误差滤波器的三维块匹配滤波方法，该方法在对高分影像的纹理丰富区进行去噪时能显著保持影像的细节信息；另外，针对单波段全极化、多波段全极化影像分别提出了一种迭代的非局部均值算法，该类算法在明显滤除全极化影像低纹理区和同质区噪声的同时，能有效保留影像的边缘结构。根据已有研究成果，借助衡量影像场景纹理复杂性的异质性指数，实现了影像纹理丰富区、同质区及过渡区域的自动确定；在此基础上，通过结合上述两类算法在处理不同场景类型影像的优势，达到了高分辨率全极化影像去噪与细节保持的优化处理。本项研究为高分全极化遥感影像去噪提供了新的研究思路与处理方法，显著提升了高分影像的质量，增强了基于极化雷达数据的解译结果和应用成果的可靠性。