基于动态纹理表达的SAR影像建模与信息提取

SAR系统正呈现多通道、多观测角、高时相、高分辨率的发展态势，而当前SAR“同质异谱”现象及单景处理模式则严重阻碍了高精度大规模SAR解译。研究表明极化SAR地物散射特性随极化基的改变存在动态规律，而时间序列、视频SAR数据形式对目标解译构成了新挑战。本项目聚焦于单景/时序/视频SAR数据信息动态化表达与建模问题，提出基于动态纹理表达的SAR影像建模新思路。利用动态纹理模型刻画极化SAR数据的“角空联合”动态极化响应特征，以动态视角发掘SAR地物散射机理；对时间序列SAR影像中包含的随机性变化、自然性变化和目标变化，研究基于生成对抗网络和“时序变化”动态表征的地物变化信息提取方法；对视频SAR数据动目标检测跟踪问题，以动态系统模型刻画不同类型纹理“时空联合运动”模式，实现动目标和背景变化的统一表达建模。本项目可为极化SAR影像散射机理动态表达和时间序列SAR数据动态解译提供新的理论和方法。

Synthetic aperture radar (SAR) systems are developing at a high speed with a trend of multi-channels, multi-view angles, multi-time phases and high resolutions. However, the phenomenon of "homogeneous region with different spectrum”, along with the single scene processing strategy seriously hinder the high-precision interpretation of time series large-scale SAR images. Recent studies show that there is an inherent dynamic change of the scattering characteristics of targets in polarimetric SAR (PolSAR) images with the change of the polarimetric basis. Similarly, both the time series SAR images and the new system SAR data contain abundant ground information and dynamic change information, and both pose new challenges for describing target information in a dynamic manner. This project focuses on the dynamic expression and modeling methods of single scene PolSAR image and time series SAR images, and puts forward a new strategy for SAR image data modeling and analysis by dynamic texture expression. We utilize the dynamic texture model to describe the joint dynamic variation trend of polarimetric response in the angle axes and the spatial axes. In this way, rich information of polarimetric SAR data can be mined and can be adopted to achieve better image interpretation results. For time series SAR data, due to the natural changes caused by seasonal differences, target changes caused by imaging parameters, and random changes caused by imaging mechanism, the SAR images contains complicated dynamic textures in both temporal axes and spatial axes, and these changes show different pattern. Describing different temporal-spatial texture mode with generative adversarial networks and dynamic system, accurate and effective ground change monitoring can be realized. For the detection and tracking of moving targets in video SAR data, the dynamic system model is used to describe the spatial and temporal joint motion mode to achieve the unified expression and modeling analysis of the dynamic changes of moving targets and background. This project can provide new theories and methods for the dynamic expression of SAR polarization mechanism and the dynamic interpretation of time series SAR imagery.