基于极化散射机理和数据统计建模的单视极化SAR图像分类方法

As one of the most important research topics in Polarimetric SAR (PolSAR) data interpretation, classification has been widely adopted in both military and civil fields. A number of PolSAR classification algorithms have been proposed, but most of them are based on multi-look data, instead of classifying single-look data directly. A filtering is required to convert single-look data to multi-look data, which could degrade the data resolution, erase image details such as edges or textures, and produce mixtures of targets. In order to extract data features from high resolution PolSAR images precisely and get rid of the potential troubles in classification of multi-look data, classification of single-look data will be investigated by exploiting physical scattering mechanism and statistical modeling of PolSAR data. First of all, statistical analysis of single-look data will be performed based on the l2-norm of scattering vectors. And then classification is carried out in two steps: urban area extraction using coherent polarimetric decomposition, and subclassification of non-urban area using exact statistical models. The study could solve the problem of statistical analysis of high resolution single-look PolSAR data theoretically and technically, as well as offer new ideas and method in PolSAR data classification.

图像分类作为极化SAR数据解译研究的重要内容之一，在军事和民用领域有着极其广泛的应用，迄今积累了许多极化SAR图像分类方法。然而大部分方法都基于多视极化SAR数据，并非直接对极化SAR系统所测得的单视数据进行分类。从单视数据到多视数据，需经过多视处理，该过程会降低图像分辨率、模糊图像边界和细节、并产生大量混合像素。为了充分提取高分辨率极化SAR图像的特征信息，克服多视极化SAR图像分类方法的局限性，本项目将研究基于散射向量二阶范数的统计分析方法，通过模型拟合优度检验等手段为高分辨率单视极化SAR数据选择精确的相干斑统计模型，并在此基础上，结合极化散射机理实现单视极化SAR图像分类。具体分类过程包括两步：通过相干极化分解进行城区提取；通过精确的统计模型实现非城区细分。本项目的研究将从理论上和技术上解决高分辨率单视极化SAR图像的精确统计分析问题，为极化SAR图像分类的发展提供新的方法和思路。

本项目重点研究了基于散射向量二阶范数的单视极化SAR数据统计分析方法，对比了不同模型下不同阶数的对数累积量，该方法可用于高分辨率单视极化SAR数据的相干斑统计模型择优。在此基础上，设计了基于统计特征和结构特征的极化SAR超像素分割方法，能够自适应的在匀质区域和异质区域生成合适大小的超像素，在保持分割边界贴近图像边缘及线状目标的同时保持分割边界的平滑。此外，还研究了利用多源遥感数据和云计算平台进行林地分类的方法，探索了光学遥感数据的多光谱信息、木材经济林的轮伐周期、SAR数据的统计特性在林地分类中的应用。在本项目的支持下，发表了3篇SCI论文，参加了IEEE IGARSS 2019、APSAR 2019等国际学术交流会议，培养了博士1名，硕士3名。