高分辨率SAR图像目标认知模型及高效算法

Comparing to the rapid development of synthetic aperture radar (SAR) technology itself, there is still no revolutionary breakthrough in application mechanism and processing methods for quantitative extraction and precise interpretation of the electromagnetic information embedded in SAR image or SAR signal, and thus there are still many problems in accurate judge of ground environment and target characteristics. In the current application areas, it basically starts from the image structure and texture features and uses the image processing method to carry out the identification of ground surface change and target characteristics. From SAR imaging principle, the project investigates the electromagnetic information of volume target, and the interaction mechanisms between volume targets and their various background environments. It also investigates the combined effects of the synthetic imaging process to the extraction of the electromagnetic scattering information of ground surface target features. Based on the above research results, new models and methods are then developed dedicated to cognize ground surface target. Finally, an application-driven, proof of concept system, using advanced mode, high-resolution SAR data to cognize target, is developed. Experiments are carried out to validate its application performance in cognizing targets at various environments. Through the project, it could innovate new ways for SAR data application, make a breakthrough in target cognition principles and methods of high-resolution SAR, achieve a substantial increase in SAR practical application effectiveness.

与合成孔径雷达(SAR)成像技术本身快速发展比较，SAR图像或SAR信号所包含的目标特征电磁信息的量化提取、精准解译从而对地物场景和目标特性的准确判断，在应用机理和处理方法上仍然没有革命性的突破。在可见的应用领域中基本上还是从图像结构和纹理特征入手，借用图像处理的方法，开展对地物变化和目标的识别。本项目将从SAR成像原理出发，深入研究体目标特征电磁信息及其与场景环境相互作用机理，研究合成成像过程对地物目标电磁散射信息的综合影响，发展高分辨率SAR一体化的信号处理模式及特征信息精准提取方法，构建SAR地物目标认知的新模型、新方法，最后通过搭建应用驱动的新体制高分辨率SAR目标认知演示验证系统，演示验证不同场景下目标识别应用效能。本项目的研究，将创新SAR应用的新途径，实现高分辨率SAR目标认知在原理和方法上的突破，大幅提高SAR的实用效益。

本项目针对传统借鉴光学图像处理方式的SAR目标认知解译模型的不足，将认知解译工作向SAR特殊成像机理端推进，在充分考虑SAR成像机理和特殊表征方式、融合人工智能方法的基础上构建目标解译模型，将目标与环境电磁散射机理、SAR固有特征获取、目标认知解译模型贯穿为有机整体，推动SAR目标认知解译理论与方法的发展。取得的重要结果和关键数据如下：.1）在目标与环境电磁散射机理与方法方面，完成了海面几何建模、舰船尾迹、地物散射建模，分别提出了基于面元生长的隐藏面快速（秒级）消除算法、基于osgEarth的三维可视化方法，基于BART算法和MPA算法实现了飞机、坦克、船只等单一体目标场景以及百万平方公里级复杂场景的SAR图像仿真，并形成了相应的软件。.2）在SAR信息提取一体化特征获取方面，围绕7类特征（大斜视成像、小入射角成像、多角度特征增强、体目标特征提取、目标稀疏特征表达、目标运动参数提取、目标图像级分辨率评估）的精确获取开展了创新研究工作，通过特征提取算法的创新，实现了大斜视二维和三维成像、典型目标多角度特征增强、稀疏特征优化表征、目标运动信息提取、图像分辨能力客观估计等，促进了SAR目标特征的理解，提升了SAR目标特征提取的多样性。.3）在SAR目标认知模型与图像解译方面，围绕4类目标（海上目标、车辆目标、建筑目标、地物目标）开展了认知解译开展研究。将深度学习算法引入至认知模型的构建中，并采取了针对SAR特点的适应性改造，提出了17种解译模型，提升了海上目标、车辆目标、建筑目标、地物目标的认知解译精度。.4）在目标认知真实性检验方面，构建了基于仿真数据、实验室成像系统数据、机载SAR数据、星载SAR数据的验证数据集，开展了认知雷达系统设计，并对本项目提出的算法开展了验证。