【作者】 李涛；

【导师】 刘国祥；

【作者基本信息】 西南交通大学 ， 摄影测量与遥感， 2014， 博士

【摘要】 地面沉降是一种危害极大的工程地质现象。地面沉降能够造成公共设施损坏,资源开发受限,严重时甚至会引起海水倒灌,从而导致土壤和地下水盐碱化,港湾设施失效,近海区域被海水淹没等。现阶段较为常用的地面沉降监测方法包括传统监测方法,全球定位系统(Global Positioning System, GPS)监测方法以及合成孔径雷达差分干涉(Differential Interferometric Synthetic Aperture Radar, DInSAR)监测方法。传统监测方法多为水准测量,精度较高,然而无法进行大规模高密度监测。GPS可以提供大规模监测方案,然而点位布设成本很高,因此难以提供高密度形变场。DInSAR很好的克服了上述两种方法的缺陷,它能够提供大范围,高密度的测量点,从而获取研究区域的高分辨率形变场,因此DInSAR正在被越来越广泛的应用于地面沉降监测之中。然而,DInSAR面临诸多应用缺陷。首先,DInSAR很容易受到时空失相干的影响。时空失相干会导致信号无法进行干涉,从而无法提供有效的形变信息。其次,DInSAR的精度受制于相位解缠的精度。相位解缠只能在高相干区域才能给出稳健的结果,低相干区域以及不连续区域会导致相位解缠结果偏离真实值,无法给出可靠的解缠相位。第三,DInSAR的处理过程较为复杂,处理过程会引入各种解算误差。这些误差可能来源于影像配准,轨道误差,大气影响等等。因此,针对DlnSAR的缺陷,有学者提出多时相雷达干涉(Multi-Temporal InSAR, MTInSAR)算法,用以获取更为可靠的形变场信息。MTInSAR中最为典型的是永久散射体雷达干涉(Persistent Scatterer InSAR, PSI)。PSI只针对影像序列中的永久散射体(Persistent Scatterer. PS)点进行分析,分析过程中,为了保证PS点的可靠性,至少应使用30景影像。由于PS点在较长的时间范围内能够保持较好的相干性,受到时空失相干和噪声影响较小,因此能够为形变场提供可靠的观测结果。一些典型的PS包括路灯、石头、高压电塔、墙体二面角等,这些地物对应的PS点在影像中占据的大小不超过一个像元。然而,由于其后向散射系数较大,其回波信号占据影像回波信号的绝大部分,从而有效的抑制了像元内噪声分量,提高了PS点的信噪比。为了获取PS点的形变时间序列,计算过程中首先需要获取点位的相对形变速率,随后将相对形变速率转变为绝对形变速率,最后通过对残差进行时空滤波计算每个点的非线性形变分量。最终的形变时间序列包含线性部分和非线性部分。PSI能够在保留DInSAR大范围监测特征的基础上,提高DInSAR的观测精度。有研究表明,在地面点信噪比足够高的情况下,PSI可以提供亚毫米级形变结果。这为工程应用提供了可靠的理论基础。虽然PSI能够提供良好的沉降观测结果,但是PS点的分布较为稀疏,相比于DInSAR来说,观测点的密度很低,无法为形变场提供足够丰富的形变细节信息。且有研究表明,在一般情况下,90%左右的地面覆盖均为面散射体,也称为分布式散射体(Distributed Scatterer, DS)。进一步研究显示,DS的物理尺寸较大,在影像中占据多个像元。虽然影像中的DS点也会受到时空失相干的影响,但是其中依然存在高信噪比的点,这些点能够为形变场提供可靠的观测结果。一些典型的DS包括道路、房顶、裸地、闲置耕地、荒山、沙漠等。这些DS对应的像素点虽然会受到时空失相干的影响,但是依然拥有中等相干性,在时间序列上能够保持相对稳定。SqueeSARTM技术可以用来进行DS点探测,探测结果表明,相对于PSInSAR向TM来说,DS点的密度可以提高至少4倍。在市郊区域,DS点的密度甚至会比PS点的密度高10倍以上。然而,SqueeSARTM将S点和DS点进行联合解算,解算过程中高质量的PS点会因为误差传播而受到低质量DS点的影响,无法保证解算结果的可靠性。因此本论文提出点目标分层分析算法,用以进行DS点探测和解算。点目标分层分析算法按照振幅离差指数(Amplitude Dispersion Index, ADI)的大小,将影像中所有的点目标分层,对每一层的像素进行逐点分析,以获取并解算PS点和DS点。解算过程中,为了保证结果的可靠性,首先需要获取PS点及其沉降参数。随后将其余的点按照ADI的步长分层,对每一层的点目标解算完毕,即可获取DS点及其形变参数。值得说明的是,为了控制点位质量,对每一层点进行解算时,都要以所有有效点目标为基础进行质量评价,只有通过质量评价的点才能作为DS点。围绕点目标分层分析算法,论文另外提出了区域增长算法和基于相位梯度的非线性形变计算方法。其中,区域增长算法可将点位的相对形变速率转变为绝对形变速率,它通过复杂的逻辑判断来替代全体点网平差算法,可以有效的减少内存的消耗,并提高解算效率和精度。模拟实验表明,在噪声较大时,区域增长算法依然可以提供形变场的整体形变信息,其表现要优于最小二乘平差算法。而基于相位梯度的非线性形变计算方法不再进行相位解缠,它使用相位梯度进行时空滤波,获取了非线性形变的相位梯度之后,使用区域增长法提供的积分路径沿路径积分,即可得到每个点的非线性形变量值。这种计算方式可以减少相位解缠过程中带来的误差,以保证形变结果的可靠性。论文使用天津市西青区的TerraSAR-X影像进行了实验,并采用同时期的水准数据进行了对比验证。验证结果表明,点目标分层分析算法提供的沉降速率均方根误差为2.5 mm/yr。同时,实验采用了基于相位梯度的非线性形变计算方法,从而获取了各沉降段的累计沉降量。验证结果表明,累计沉降量的均方根误差为3.8 mm,达到三等水准测量的精度要求。对比结果证明了论文提出的算法的可靠性。此外,论文还使用了上海市宝山区以及香港国际机场的TerraSAR-X影像进行了实验,实验表明,论文提出的算法不仅可以为上海地铁沿线沉降提供详细的形变信息；而且可以为香港国际机场地表沉降提供详细的形变信息,且形变信息分布特征与机场的地质特征具有很高的一致性。实验结果证明了论文提出的算法的有效性,即算法可以获取研究区域的高空间分辨率形变场。论文为MTInSAR技术的工程应用提供了良好的实验依据。更多还原

【Abstract】 Land subsidence is a destructive engineering geologic phenomenon. It can cause damage to public facilities, resource development, and even introduce indwelling which is responsible for salinization of soil and fresh water, invalidation of harbor facilities. as well as submergence of coastal areas. Three main tools can be used to detect land subsidence. They are traditional method such as leveling measurement, global positioning system (GPS). and differential interferometric synthetic aperture radar (DInSAR). Leveling measurement possesses the highest accuracy but the lowest spatial density of the measurement points. GPS is capable of providing measurement points in large spatial scale. However. the cost is too high to afford a high density for deformation field. DInSAR. which provides large spatial scale and high density of measurement points, has perfectly overcome the shortcomings of the aforementioned two technologies. Deformation field with high spatial resolution is able to be extract from the DInSAR technology. Therefore, DInSAR is used more and more frequently in land subsidence monitoring.However. DInSAR is limited by some technical flaws. First. DInSAR is sensitive to spatio-temporal decoherence. The spatio-temporal decoherence makes the signals not able to interfer. Therefore, it is difficult to detect deformation information in low coherent areas. Second, accuracy of DInSAR is limited to accuracy of phase unwrapping algorithms. Phase unwrapping is not robust in low coherent areas. The results of those areas are therefore biased and not reliable. They can’t be used for in-depth analysis. Third, procedures of DInSAR are so complex that errors in the final results may originate from any step of the procedures. The errors may be caused by the image coregistration, orbital error, and atmospheric phase screen, et al. Therefore some scholars have proposed the multi-temporal InSAR (MTInSAR) to provide more reliable deformation information.Persistent scatterer InSAR (PSI) is one of the most popular MTInSAR methodologies. PSI concentrates on the persistent scatterer (PS) instead of all kinds of the ground coverage. At least 30 SAR images are suggested to use to make the PS pixels statistically reliable. Because the PS pixels keep coherent in a long time, with little effects of spatio-temporal decoherence and noise components, they are reliable in providing reliable results for deformation field. Some typical PSs include street lamps, rocks, electricity pylons, and dihedral corners in residential regions. Although the sizes of the PSs in SAR images are less than one pixel, the backscattering of the PSs occupies most of the backscattering information of the sub-elements within a single pixel, bringing high signal-to-noise ratio (SNR) by suppressing the noise components. The standard procedures of PSI follow two main steps. First, the relative deformation parameters have to be calculated. Then the relative deformation parameters have to be converted to the absolute ones with respect to a common reference point. Finally, the nonlinear deformation components are extracted using spatio-temporal filtering. The final deformation time series are calculated by adding the nonlinear deformation components to the linear ones. By following the procedures of PSI, the precision is increased compared to InSAR. Besides, the character of large spatial scale of measurement points is kept in PSI. In-depth analysis shows that the accuracy of PSI can reach up to submillimeter level if the SNR of PS pixels are high enough. This is an encouraging conclusion which provides authentic basis for engineering application of PSI.PSI provides high reliability to the deformation information. However, the density of PS pixels is lower than that of InSAR. Therefore. spatial resolution of deformation field is too low to provide deformation details. It is reported that 90% of the ground coverage is distributed scatterer (DS). Therefore, DS is potential in supplying deformation details in MTInSAR analysis. Unlike PSs. DSs are large in physical sizes. They occupy tens of pixels in SAR images. They are affected by spatio-temporal decoherence, but pixels with high SNR can also be detected among DS pixels. Those pixels are useful in providing reliable deformation information. Some typical DSs include road, roofs, bare lands, idle farmlands, waste mountains, desserts, et al. Although the DS pixels are affected by spatiotemporal decoherence, they still have moderate coherence values. They are stable in time series. A typical DS detection methodology is SqueeSARTM which obtains at least a 4 times larger density than PSInSARTM. At the suburban areas, the density can be as large as 10 times. However, SqueeSARTM calculates both PS and DS pixels jointly. Error propagation from DS pixels to PS pixels lowers the quality of the latter ones. Thus the results cannot be ensured. Therefore in this thesis we propose a hierarchical point analysis (HPA) methodology to detect the DS pixels and calculate the corresponding deformation parameters.The HPA uses amplitude dispersion index (ADI) to classify all the pixels to several groups. Pixel-by-pixel strategy is applied to analyze each of the pixels, thus extracting both PS and DS pixels. To make the result authentic, the PS pixels are first calculated. Then the groups are maintained by considering ADI intervals. And DS pixels are extracted after treating the groups one by one. It is worth stating that each DS candidate is assessed by using the valid pixels extracted from the previous groups. Only those that have passed the quality assessments are maintained as DS pixels.Besides, the thesis proposes a region growing algorithm and phase gradient-based nonlinear deformation components extraction method to accomplish the processing chains of the MTInSAR. The region growing method is used to convert the relative deformation parameters to the absolute ones. This method is applied to conduct kinds of logical judgments to take place of least square adjustment. It is efficient in providing high accuracy. And low computing resources are occupied by comparing to least square adjustment. The phase gradient-based nonlinear deformation components extraction method applies no phase unwrapping algorithm. It uses the phase gradient to do spatio-temporal filtering and get the nonlinear deformation gradients on arcs. Thereafter, the integral paths maintained after region growing are used to provide integration for those gradients and get the deformation values for each point. This method is reliable by avoiding phase unwrapping errors.Finally, the TerraSAR-X images acquired over Xiqing district, Tianjin. China is applied to validate the algorithms proposed in the thesis. Meanwhile, the leveling data obtained during the same observation period are introduced for comparison purpose. Experimental results show that the root-mean-square error of subsidence rates provided by HPA is 2.5 mm/yr. The nonlinear subsidence components are extracted using the phase-gradient based algorithm, thus calculating the subsidence time series. The subsidence time series are compared to the corresponding leveling results. Results show that the RMSE of accumulated subsidence values are 3.8 mm. Those are evidences that the proposed methodologies are validated in providing accurate deformation information. Besides, the thesis has conducted experiments by using the TerraSAR-X images acquired from Baoshan district, Shanghai as well as Hong Kong International Airport (HKIA). Experimental results show that the proposed methodologies are able to provide detailed deformation information along the railways in Baoshan district, Shanghai. Besides, the detailed deformation information provided for HKIA shows high consistency with the geology information in the study area. Those are the evidences that the proposed methodologies are validated in providing high spatial resolution for the useful measurements. The proposed MTInSAR algorithms are expected to facilitate deformation monitoring in engineering applications.更多还原