【作者】 罗小军；

【导师】 黄丁发； 刘国祥；

【作者基本信息】 西南交通大学 ， 大地测量学与测量工程， 2007， 博士

【摘要】 合成孔径雷达差分干涉技术是新近发展起来的用于监测大范围地表形变的新技术，它具有精度高、视域广等特点，可与基于点观测的GPS和水准测量形成优势互补，为形变监测和地球物理研究提供一种有效的空间对地观测新途径。但是，合成孔径雷达差分干涉技术的应用受到失相关和大气延迟两大因素的制约。近年发展起来的基于永久散射体的雷达差分干涉技术是目前克服失相关和大气延迟的最有效方法。然而，该方法在一些关键数据处理方面(永久散射体探测、建网、模型解算)还存在问题。为此，本文对永久散射体雷达差分干涉技术理论进行系统研究，对其中的关键算法做重点研究，探索解决该方法存在问题的新途径，以提高永久散射体雷达差分干涉技术探测地表形变的精度和可靠性。本文首先从干涉相位分解的角度出发，分析了地形、形变、大气延迟和失相关噪声对干涉相位的贡献，阐明了合成孔径雷达差分干涉探测地表形变的原理，为PS-DInSAR差分相位建模奠定了基础。SAR影像精确配准是PS-DInSAR的基础和关键步骤之一，论文通过对三种SAR影像配准算法(相干系数法、相位差影像平均波动函数法、最大频谱法)的比较研究，得出相干系数法的配准质量最好。在PS-DInSAR中，可采用相干系数法，并借助Doris软件实现多幅时序SAR影像的配准。为提高PS识别的准确性与可靠性，本文提出了探测PS的新方法—振幅信息双阈值法。该方法既考虑了永久散射体散射的稳定性，又考虑了其回波信号的高信噪比特性。通过与其它方法的对比和对上海城市地区26幅ERS-1／2卫星SAR影像的PS探测实验，证实了振幅信息双阈值法探测的PS有效而且更加可靠。在PS邻域差分相位建模中，首次提出采用PS三维建网方法确定PS的邻域关系。并且，在根据PS三维网络求解得到PS邻域差分值之后，采用间接平差法解决了PS三维网的几何矛盾，获得了各PS点的线性形变速率和高程误差，并实现了各PS点差分干涉相位的解缠。同时引入稳健估计方法抑制在平差过程中可能存在的粗差的影响。对上海陆家嘴地面沉降的研究证实，PS三维建网方法获得的结果精度优于基于像平面坐标的PS二维建网方法获得的结果精度。在PS邻域差分相位模型参数估计中，提出了解空间搜索法。在依据先验知识确定解空间的大小、位置和搜索策略后，可较快地获得参数的最佳估计值。同时对主影像和从影像大气相位的估计方法作了改进，在原有估计方法中增加了低通滤波处理步骤，这样可消除大气相位中可能存在的失相关噪声的影响。对上海陆家嘴地面沉降的研究证实了解空间搜索法能有效、可靠地估计PS邻域差分相位模型参数，改进的大气相位估计方法能有效削弱大气相位中的失相关噪声。最后，应用基于Matlab环境编写的一套PS-DInSAR处理程序，成功探测了上海陆家嘴地区在1992～2002年间的地面沉降，获得陆家嘴地区在10年内的最大形变量为183.2mm，最小为78.4mm，总体平均形变速率为13.72mm／a，与实测结果具有很高的一致性。这表明利用PS-DInSAR技术探测地表形变是可行的，本次研究结果以及相应的算法和处理程序是有效而可靠的。此外，对形变和大气相位的时空相关性做了统计分析，得出形变在时间上和小于4km的空间范围内具有很强的相关性；大气延迟在小于2km的空间范围内具有强相关性，而在时间上却不具有相关性。更多还原

【Abstract】 Differential synthetic aperture radar interferometry(DInSAR) is a newly developed technique for monitoring large-scale ground deformation with some prominent advantages such as high accuracy and pantoscopic view. It therefore can greatly complement many conventional point-based geodetic techniques such as GPS and leveling. This provides a viable space-geodetic approach for ground deformation detection and geophysical studies. However, decorrelations and atmospheric delays impede the applications of DInSAR. The newly-proposed idea, called DInSAR based on permanent scatters (PS-DInSAR), is at present regarded as the most efficient approach in overcoming both decorrelations and atmospheric delays. Nevertheless some problems still exist in the key processing procedures of PS-DInSAR such as PS detection, network construction and model estimation. Thereby, this thesis focuses on investigating basic principles of PS-DInSAR and critical algorithms as well as exploring new approaches so as to improve both accuracy and reliability in PS-DInSAR.The influence of terrain, deformation, atmospheric delays and decorrelations onto interferometric phases is first discussed by analyzing the components of interferometric phases. The principle of DInSAR for deformation detection has been described based on the analysis of phase components. This investigation laid a foundation for modeling differential phases in PS-DInSAR.The accurate co-registration of SAR images is one of key procedures in data processing of PS-DInSAR. The conclusion that the performance of correlation-coefficient method (CCM) ranks first has been drawn by comparing the three accurate co-registration algorithms such as CCM, method based on the average fluctuation function of phase-difference image (MAF), as well as maximum-spectrum method (MSM). CCM is therefore chosen as the approach for co-registering time-series SAR images in PS-DInSAR and Doris software can be utilized for this purpose.To improve accuracy and reliability in identifying PSs, a new PS detection algorithm, called a dual-threshold method based on amplitude information, has been proposed. It considers both PS’s temporal stableness of radar backscattering and its high signal noise ratio (SNR) of radar echoes. This method has been proven effective and more reliable by comparing with other approaches by means of the experiments with 26 ERS-1/2 SAR images of Shanghai.For modeling of PS neighborhood differential phases, it is the first time that the method of creating three-dimensional (3D) PS network is proposed to more precisely determine the neighborhood of PSs. After deriving differential values along arcs in the network, a parametric adjustment method is used to eliminate geometric inconsistency of the 3D network, and thus estimating the terrain error and linear deformation velocity, as well as unwrapping differential phases of PSs. A robust estimation method is also adopted to avoid contamination of possible gross errors during the network adjustment. The investigation on detecting ground subsidence over Lujiazui of Shanghai shows that the results derived from 3D PS network are more accurate than those derived with 2D PS, network created with image planar coordinate system.A new algorithm, called solution-space search (SSS), has been proposed to estimate the parameters in the model of PS neighborhood differential phases. After determining solution-space size, location and search strategy according to some apriori information, the SSS method can be used to compute the optimal parameters rapidly. Besides, the atmospheric phase estimation algorithms for main and slave images have been improved by adding the low pass filtering procedure to raw processing steps. The experiments on detecting ground subsidence over Lujiazui in Shanghai have proven that the SSS method can effectively and reliably estimate the parameters in the model of PS neighborhood differential phases, and the improved algorithms of estimating atmospheric phases are able to mitigate the imapcts of decorrelation noises. Finally, the ground subsidence from 1992 to 2002 over Lujazui area of Shanghai has been successfully detected using a set of PS-DInSAR computer programs which is developed in the Matlab environment. It is found that the maximum and minimum subsidence over this area within 10 years is 183.2mm and 78.4mm, repectively, and the averaged displacement velocity is 13.72mm/a. The results are in good agreement with the ground-based measurements. This indicates that the ground deformations can be efficiently detected and tracked with PS- DInSAR, the algorithms and computer programs developed in this thesis are viable and reliable. Moreover, the statistical analysis on temporal and spatial correlation in terms of deformation and atmospheric phase has been conducted. It can be concluded that the deformations exhibit strong temporal correlation as well as strong spatial correlation within 4km, and the atmospheric delays show strong spatial correlation within 2km without temporal correlation.更多还原