基于L1范数正则化的大场景实时稀疏SAR成像方法研究

In order to solve the existing challenges and key problems in the design of sparse synthetic aperture radar (SAR) imaging system, especially the problem of real-time data processing, this project focuses on the research about the L1-norm regularization based real-time sparse SAR imaging technique for large-scale scenes. To solve the problem about the huge computational cost of the conventional observation matrix based sparse SAR imaging method, a novel real-time model for sparse SAR imaging will be constructed in this project. This model uses a two-dimensional operator to replace the observation matrix to reduce the memory requirement in the sparse recovery. For the challenge of real-time sparse reconstruction of large-scale scenes, the L1-norm regularization based real-time sparse SAR imaging method will be explored to achieve the high-resolution real-time recovery of large-scale sparse and non-sparse scenes. Now the existing typical regularization recovery algorithms in sparse SAR imaging can obtain a sparse estimation of the considered scene only, which does not preserve the image phase and statistical distribution. To solve this problem, this project will propose a novel thresholding based L1-norm regularization recovery algorithm, and apply it for the real-time sparse SAR imaging. The solution of this algorithm not only protrudes the target as recovered image of existing algorithms, but also has the same phase as and similar background statistical distribution to matched filtering reconstructed image only with the decreased amplitude. Finally, the experiments based on simulated, real airborne and spaceborne data are used to validate the proposed real-time sparse SAR imaging method, so as to complete the theory and processing technique of sparse SAR imaging, show its potential, and hence provides the theoretical and technical support for the development and further application of the novel sparse SAR system.

面向现有稀疏SAR成像系统无法进行实时数据处理等难点和关键性问题，本项目拟开展基于L1范数正则化的大场景实时稀疏SAR成像方法研究。针对基于现有模型的稀疏SAR成像方法计算代价大的问题，构建一种实时稀疏成像模型，使用二维运算算子替代观测矩阵以减少对模型所需存储空间要求；针对大观测场景的实时稀疏成像问题，探索基于L1范数正则化的实时稀疏SAR成像方法，实现对任意大观测场景的高分辨率实时重建；针对现有稀疏SAR成像正则化重构算法恢复图像相位信息及统计特性破坏严重的问题，设计一种基于阈值收缩的全新L1范数正则化重构算法，并将其用于大观测场景的实时成像，使得重建图像较匹配滤波结果具有更优质量的同时，亦能很好保持相位及统计特性。最后通过机载、星载数据进行实验验证，完善所提出的模型和处理方法，充分发挥稀疏SAR成像潜力，为新一代稀疏SAR成像系统的研制和应用提供技术支撑。

本项目研究面向高分辨率对地观测、微波遥感系统发展等重要应用需求，拟开展稀疏SAR快速成像技术的研究工作，突破稀疏SAR成像雷达工程化实现的算法瓶颈，为高分宽幅星载稀疏SAR系统的研制提供算法支持。. 项目针对稀疏SAR成像雷达工程化实现中所面临的算法计算复杂度高、大场景实时重建困难、稀疏重构图像相位信息及统计特性破坏严重三个难点和关键性挑战，构建了基于正则化的稀疏SAR实时数据处理框架：1）提出了全新的实时稀疏SAR成像模型，显著减少了稀疏成像对存储空间的要求；2）提出了基于L1范数正则化的实时稀疏SAR成像方法，实现了大观测场景的高分辨率实时重建，计算时间降至经典匹配滤波方法3倍以内；3）提出了用于解决L1范数正则化问题的全新重构算法，并将其引入稀疏成像，使得稀疏重构图像在提升质量的同时，有效保持了相位信息及统计分布。项目创新成果为新一代星载稀疏SAR成像雷达的设计和研制提供了算法支撑。. 本项目完成了合同要求的各项任务指标，具体研究成果如下：. 在国内外期刊和会议上发表论文17篇，包含IEEE Transactions on Geoscience and Remote Sensing、IEEE Geoscience and Remote Sensing Letters、IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing等IEEE高水平期刊论文8篇；在IGARSS等国际会议上发表论文6篇；申请发明专利5项。项目负责人获得“全国青年岗位能手”称号，入选国防科技领域青年托举人才计划，培养博士研究生2名、硕士研究生6名。