天基干涉相位图像智能快速降噪与解缠技术研究

Interferometric phase images acquired from the spaceborne interferometric synthetic aperture radar are widely used in getting the topography of the earth surface. Because of the measuring instruments and environment, the measurements in the phase imaging systems are noisy and usually 2π-periodic functions of the true phase. The phase images are also of large scale which would cost amount of memory and time when processed by computer. In our research, we focus on the denoising and unwrapping algorithm of the interferometric phase images based on modern intelligent algorithms to improve the accuracy of the measurement and the processing speed. We analyze the noise characters in the phase images, and would propose denoising algorithms based on the constraint to sparse codings of the patches in the same cluster and adaptive dictionary learning in the complex valued domain of the phase images. The purpose is to remove the noise and retain the details as more as possible in the phase images. In order to unwrap the phase rapidly without the guidance of quality map, we would model the unwrapping problem to an optimization of the Markov Random Fields problem and research on the solution of the problem. Finally, based on the denoising and unwrapping algorithms proposed above, we would also research on the fast algorithm of the large scale phase images by utilizing the parallel computing of GPU.

基于天基合成孔径雷达获取的干涉相位图像广泛地用于地形拓扑的生成等应用中。由于受到环境和测量设备的影响，相位图像系统中获取的测量结果通常含有噪声，而且是真实相位值的模2π值。通过天基合成孔径雷达获取的干涉相位图像还具有大规模的特点，在处理时会耗费大量时间和计算机内存。在本研究中，重点关注运用最新智能算法解决干涉相位图像的降噪和解缠问题以提高干涉测量结果的精度和处理速度。首先分析相位图像中的噪声特点以及自相似性，研究复数域中基于聚类相似稀疏编码和自适应词典学习的降噪方法，在移除噪声干扰的同时尽可能保留相位图像中的细节信息。其次，通过将相位解缠问题建模为马尔科夫随机场的最优化问题，并且研究该问题的快速求解方法来降低解缠对质量图的依赖性。最后，基于所研究的降噪和解缠算法，利用GPU的并行特性，研究大规模相位图像的降噪和解缠快速算法，提高处理效率。

按照项目进度安排与申请书中研究内容的要求，本课题针对提高干涉测量中的精度、准度和速度问题开展了相关研究，有效解决了测量噪声导致的精度降低、真实相位的模2π值不准确测量和大规模天基干涉相位的解缠速度慢等问题。研究基于干涉测量的噪声特点，通过提出复数域的基于变量分裂的在线字典训练算法和分组稀疏编码算法，有效提高了对于干涉相位图像中的噪声抑制问题，在移除噪声干扰的同时尽可能保留相位图像中的细节信息；研究针对相位缠绕的特点，提出运用马尔科夫随机场对相位解缠问题进行建模，并且研究提出了一种基于改进二阶伪布尔最优化算法进行问题求解，该解缠算法能够降低解缠对质量图的依赖性，进而提高了相位解缠的适应性。最后研究针对大规模天基干涉相位的解缠问题，研究提出基于分割与拼合的相位降噪解缠并行算法，有效降低了算法的时间复杂度。研究中还将提出的算法成果应用到了高光谱降噪解混中，支撑多名硕士博士开展论文研究。项目相关研究成果发表论文16篇，其中SCI检索7篇，申请国家专利1项，国防专利1项。