基于物理模型的极化SAR自动目标识别研究

Target detection under complex environment and target recognition under extended operating conditions are still two difficult issues in synthetic aperture radar (SAR) automatic target recognition. In order to resolve these two issues, this project will make use of the high-resolution fully polarimetric SAR (PolSAR) images. Starting from two kinds of physical models, i.e., the physical scattering mechanism model and the scattering center model, the project will be performed from the following three aspects. Firstly, based on the parametric scattering mechanism model set and the idea of sparse reconstruction, the polarimetric target decomposition with the capability of model adaptation will be studied. It will be used to analyze the scattering mechanisms of targets and clutters, which provide the basis for the following study on target detection and recognition. Secondly, by analyzing the physical scattering mechanisms, we plan to construct the parametric model for the composite target, which includes several different elementary scattering mechanisms. Then the polarimetric test statistic will be designed by considering the composite target model. In this way, the composite target detection approach based on physical scattering mechanisms will be studied in order to detect composite targets under the complex environment. Finally, the attributed scattering center model parameter estimation will be performed by making a combined use of the sparse characteristics in both the image and the frequency domain. The polarimetric features of the scattering centers will also be extracted. Then based on attribute robust scattering centers, the polarimetric robust target recognition feature extraction will be studied, in order to accomplish the robust target recognition under extended operating conditions by combining the polarimetric information. The project will technically support the application of PolSAR data in civil economy and national defense.

针对合成孔径雷达（SAR）自动目标识别（ATR）中的两个难点问题，即复杂环境下的目标检测与扩展工作条件下的稳健目标识别，本项目将基于高分辨率全极化SAR（PolSAR）数据，从物理散射机理模型与散射中心模型两种物理模型出发，对以下三方面内容展开研究。首先，基于参数化物理散射机理模型集与稀疏重构思想，研究具有模型自适应能力的极化目标分解方法，以分析目标与杂波的散射机理，为后续的目标检测与识别奠定基础；其次，通过分析散射机理，建立包含多种基本散射类型的复合目标的参数化模型，并结合该模型构造极化检测量，以实现复杂环境下基于物理散射机理的复合目标检测；最后，拟联合图像域与频率域的稀疏性估计属性散射中心模型的物理参数，并提取散射中心极化特征，研究基于属性稳定散射中心的极化稳健识别特征提取，以结合极化信息实现扩展工作条件下的稳健识别。为PolSAR数据在国民经济与国防建设中的应用提供技术支撑。

复杂环境下的目标检测与扩展工作条件下的稳健目标识别，是SAR自动目标识别中的两个难点问题。本项目针对这两个难点问题，开展了基于物理模型的极化SAR自动目标识别方法的研究。主要研究内容和成果包括以下几个方面：（1）研究了具有模型自适应能力的极化目标分解方法，包括基于约束稀疏表示的极化相干矩阵分解方法，以及极化相似度匹配下的极化相干矩阵散射能量分解方法；（2）研究了基于物理散射机理的复合目标检测方法，包括基于超像素级散射机理分布特征的极化SAR目标检测方法、基于监督非相干字典学习的极化SAR舰船目标检测方法、以及基于低秩字典学习与稀疏表示的极化SAR目标检测方法；（3）研究了基于物理模型的极化稳健目标识别方法，包括基于稀疏表示的属性散射中心提取与参数估计方法、基于全极化属性散射中心模型的SAR目标属性特征提取方法、基于非负稀疏表示的SAR目标识别方法、联合阴影与目标区域图像的SAR目标识别方法、姿态图像缺失情况下的SAR目标识别方法、以及基于多信息字典学习与稀疏表示的SAR目标识别方法等。本项目的研究成果可为SAR自动目标识别技术及其应用的发展提供一定的参考。