面向高程反演的高分InSAR图像的复卷积稀疏表示去噪方法研究

随着新一代对地观测SAR卫星向着大尺度、高分辨率、高重访频率等方向发展，由其得到的干涉图能在大尺度上更加精细地反演出所观测地物的高程信息。然而，现有的方法在干涉图去噪的精度及尺度上存在瓶颈。本项目聚焦干涉图去噪问题，通过研究复数域卷积稀疏表示方法的去噪模型与求解算法，确立大场景、高分辨率干涉图像下面向高程反演的干涉图去噪方法框架。具体包括：1）建立复数域卷积稀疏表示的干涉图去噪模型及相应的优化求解方法；2）结合区域高程拓扑结构等先验信息来提升卷积稀疏表示去噪模型的效果；3）研究基于卷积稀疏表示的多基线干涉图去噪模型及优化方法；4）优化所提方法的时效性。本项目突破了现有的方法在大尺度、高分辨率干涉图去噪下精度的瓶颈问题，对于大范围、高精度的高程反演具有重要意义。

As the next generation of SAR systems is developed towards the directions of large-scale, high-resolution and high-revisit frequency, the obtained interferograms can reconstruct the elevations of the observed areas with the preservations of details at a large scale. However, the conventional methods exist bottlenecks in the accuracy and scalability. This project will target on the denoising problem of interferograms, and develop the mathematical models based on convolutional sparse representation and the associated optimization algorithms in complex domain. Moreover, this project will provide the parallel version of the proposed methods, which will preserve the time efficiency for large-scale image processing. Specifically, the project is composed of: 1) developing the mathematical models based on convolutional sparse representation and the associated optimization algorithms in complex domain; 2) integrating the prior knowledge, such as the topography of area elevation into the reconstruction method based on convolutional sparse representation; 3) developing the multi-baseline interferogram denoising approach based on convolutional sparse representation; 4) extending the proposed methods into parallel version which can improve the time efficiency. The research of the project aims to promote the cross-disciplinary integration of mathematics, information, and remote sensing, and to improve the accuracy and efficiency of InSAR denoising methods, which can provide the solid base for elevation reconstruction from large-scale, high-resolution interferograms.