基于改进移动最小二乘解缠相位去噪  

A Denoising Method Based on Improved MovingLeast Square Phase-Unwrapping

作　　者：许章平 刘成洲 庄艳 XU Zhangping;LIU Chengzhou;ZHUANG Yan(Yantai CIMC Raffles Offshore Engineering Co.,Ltd.,Yantai 264000,China;College of Geometics,Shandong University of Science and Technology,Qingdao 266590,China)

机构地区：[1]烟台中集来福士海洋工程有限公司,山东烟台264000 [2]山东科技大学测绘科学与工程学院,山东青岛266590

出　　处：《测绘与空间地理信息》2020年第7期111-114,共4页Geomatics & Spatial Information Technology

摘　　要：准确的相位解缠是SAR卫星监测地表变形的前提和关键,然而当待解缠相位图像中存在严重噪声、不连续或不连通区域时,当前已经提出的许多的相位解缠算法会在条纹线边缘出现2π跃变噪声。这些2π跃变噪声和干涉相位本身存在的噪声会使SAR监测图像不能准确表达地表的变形情况。为此,本文提出一种基于改进移动最小二乘滤波方法,该方法可有效地去除相位解缠后的条纹线边缘的跃变噪声和干涉相位本身存在的噪声。首先使用SNAP软件对Sentinel-1A的两组SAR影像进行差分干涉(DInSAR)处理;其次利用迭代对流层分解模型(Iterative Tropospheric Decomposition,ITD)对解缠后的图像进行大气校正预处理,去除大气相位对解缠结果的影响;最后对预处理后的解缠相位进行改进最小二乘滤波,获得最终的区域变形图。通过与移动最小二乘滤波和高斯滤波的去噪结果对比,改进最小二乘滤波,可以既保留最小二乘最佳拟合曲面的特性,在“跃变式”噪声下也能达到很好的去除效果,同时也不会将下沉“谷底”当成去噪对象。将实测水准数据作为参考,改进最小二乘滤波更为接近下沉曲线。改进移动最小二乘滤波,在继承了移动最小二乘的最佳拟合曲面特性的同时,也对“跃变”噪声具有很好的消除性。该方法仍需进一步改进:针对大面积噪声区域仍然无法做到精确去除;移动最小二乘方法针对大矩阵仍然没有好的计算方法,费时费力。Accurate phase-unwrapping is the precondition and key for SAR satellite to monitor surface deformation.However,when there are serious noises,discontinuous or disconnected regions in the phase image,many of the existing phase-unwrapping algorithms have 2πjump noise at the edge of fringe line.The noise of these 2π and noise in interference phase itself will make the SAR monitoring image fail to accurately express the surface deformation.Firstly,differential interference image of the two sets of Sentinel-1A SAR images is acquired by the SNAP software and then corrected by using the iterative tropospheric decomposition model(ITD).The effect of atmospheric phase on the result of differential inference SAR(D-InSAR)is eliminated,and the final region deformation map is obtained by applying the improved least square filtering on the unwrapped phase.By comparing the denoising result of the moving least square filter to the Gaussian filter,the proposed filter retains the characteristic of the best fitting surface of least square,removes the“Jump”noise,and avoids taking the sinking“bottom”as a denoising object.By taking the measured level data as a reference,the proposed filter is closer to the sinking curve.It not only inherits the best fitting surface characteristic of the moving least square,but also eliminates the“Jump”noise.The method needs to be improved in the following ways:It is still impossible to get accurate removal for the large area noise region;The moving least square method is still not good for large matrices,time-consuming,and laborious.

关 键 词：DINSAR 移动最小二乘 变形监测 高斯滤波 大气改正 

分 类 号：P228[天文地球—大地测量学与测量工程]