基于Bayesian位移场的SAR图像精确配准方法研究

SAR image registration is the necessary technique in various SAR applications. Registration accuracy plays an important role in the subsequent processes, such as interferometry, image fusion, change detection and so on. With the improvement of SAR resolution, geometric differences among the high resolution SAR images become obvious. Precise registration of such SAR images becomes a nontrivial task since a change in radar-acquisition geometry generates image shifts, which, in principle, depend on the topography and the Radar parameters. Due to the poor performance of global/local transformation, the traditional registration methods gradually can’t keep up with the precision requirement of SAR applications. In this project, statistical modeling mechanism is introduced for high resolution SAR image registration. Precise and fast registration is derived by assign a displacement vector to each pixel. The project can be divided into three parts. First, a basic Bayesian displacement field model is proposed to provide a unified Bayesian framework, which can be used to exploit the priori information of image. Second, the basic model is improved for SAR image registration, by considering the effect of texture, topography, surface features and so on. Finally, we infer the posterior distributions of the proposed displacement fields, and design the registration for SAR image. This project will enrich and improve the theory of SAR image registration, and the fruits of the research will exposed to a broader audience.

SAR图像配准是SAR应用不可或缺的重要技术，精确的配准结果是干涉测量、图像融合、变化检测等诸多后续环节的必要条件。随着传感器空间分辨率的提高，SAR图像之间呈现出复杂的几何差异，传统的全局/局部变换模型逐渐无法满足当今SAR应用在配准精度方面的要求。本项目旨在将Bayesian统计建模的思想引入到SAR图像的配准问题中，基于位移场模型计算每一个像素的位移矢量，实现高分辨率SAR图像的精确、快速配准。首先，建立Bayesian位移场基本模型，为发掘图像先验信息提供统一的Bayesian框架；然后，分析纹理、地形、地物等因素对图像灰度和图像几何的影响，建立适用于SAR图像的Bayesian位移场改进模型；研制统计推导算法，根据位移场的后验分布实现SAR图像精确配准。本项目的研究将有助于提高现有SAR图像配准方法的精度，为武器制导、地面沉降监测、灾害监测与评估等实际应用，提供有效的技术支持。

SAR图像配准是SAR应用不可或缺的重要技术，精确的配准结果是干涉测量、图像融合、变化检测等诸多后续环节的必要条件。随着传感器空间分辨率的提高，SAR图像之间呈现出复杂的几何差异，传统的全局/局部变换模型逐渐无法满足当今SAR应用在配准精度方面的要求。本项目将 Bayesian 统计建模的思想引入到SAR图像的配准问题中，通过位移场模型计算每一个像素的位移矢量，实现高分辨率 SAR 图像的精确、快速配准。首先，建立非参数 Bayesian位移场基本模型，为发掘图像先验信息提供统一的 Bayesian 框架；然后，分析纹理、地形、地物等因素对图像灰度和图像几何的影响，建立适用于SAR图像的Bayesian位移场改进模型；研制统计推导算法，根据位移场的后验分布实现SAR图像精确配准。.通过本项目的研究，突破了高分辨率SAR图像精确、快速配准的技术瓶颈，丰富了SAR图像精确配准理论与技术，具体表现为：建立了位移场 Bayesian 建模的理论基础；提出并验证了适用于 SAR 图像配准的Bayesian 位移场模型；研制了基于GPU 的位移场快速推导算法，为实际相关应用提供了精确、快速的配准技术。相比于现今先进配准方法，本项目提出Bayesian 位移场模型的在配准精度上和计算速度方面均有显著提升。本项目的研究成果能有效提高现有SAR图像配准方法的精度，将为武器制导、地面沉降监测、灾害监测与评估等实际应用，提供坚实的技术支持。