基于复杂结构参数化建模的高分辨率SAR图像建筑物三维重建方法研究

As an important part of Synthetic Aperture Radar (SAR) image interpretation research, building reconstruction from high resolution SAR images is theoretical significant and has a wide range of applications. Noise and interferences from nearby objects are very common in meter resolution SAR images, and it affects the precision and robustness of bottom-up building reconstruction methods. Moreover, the existing top-down methods limit the scope of their research in flat-roofed buildings. In order to simplify the problem, they obtain most of parametes of the buildings in advance, and only calculate a small part of the parametes in their research.They also have shortcomings in energy function construction, target structure representation and energy function optimization. To solve the problems aforementioned, this project focuses on the parametric modeling method of the building, the similarity measurement method between model and image data and the method of solving the energy function. We extend the research scope from a single flat-roofed object to neighborhoods containing multiple different structural objects, extend the parameter inversion range from one or a small part of parameters to all parameters, and improve the accuracy, robustness and applicability of 3D reconstruction algorithms for buildings significantly. It will promote the development of SAR image interpretation technology and improve the application level of SAR.

作为SAR图像解译领域的重要研究内容，高分辨率SAR图像建筑物三维重建研究具有重要的理论意义和广泛的应用价值。现有方法多采用自底向上的思路，高分辨率SAR图像来自噪声和附近其他物体的干扰限制了这类方法的准确度和鲁棒性，而已有的自顶向下方法将研究对象局限于单个平顶建筑物，并且为简化问题往往事先采取其他手段获得目标的多个参数，只反演目标的一个或小部分参数，存在目标结构单一、忽视区域特征和上下文语义以及参数反演困难等不足。针对以上问题，本项目着重研究了自顶向下方法中复杂结构的建模表示方法、目标模型与图像之间相似性的度量方法，以及如何更有效、快速地求解相应的优化问题等关键技术，将处理对象由单个平顶目标推广到包含多个结构不同目标的局部邻域，并将参数反演范围由一个或少部分参数扩展到全体参数，显著提高了建筑物三维重建算法的准确度、鲁棒性和适用性，以促进SAR图像解译技术的发展，提升SAR的应用水平。

本项目结合高分辨率SAR图像目标解译的应用需求，针对现有方法在建筑物三维重建中存在的不足，研究目标的建模表示、目标模型与数据的匹配度量方法和目标参数反演方法等建筑物三维重建过程中涉及的关键算法和问题，在此基础上实现自顶向下的SAR图像建筑物三维重建，并完成原型系统的构建，通过实验与分析验证相关算法的可靠性，以提高建筑物三维重建算法的准确度、鲁棒性和适用性，促进SAR图像解译技术的发展，提升SAR的应用水平。项目按计划实施了各项研究，在项目的支持下，发表SCI论文2篇，EI论文2篇。培养研究生4名，包括博士研究生2名，硕士研究生2名，完成了申请阶段的预期成果。