Back to results list
Please use this identifier to cite or link to this item:
|Title:||InSAR coherence estimation and applications to earth observation||Authors:||Jiang, Mi||Keywords:||Synthetic aperture radar.
Hong Kong Polytechnic University -- Dissertations
|Issue Date:||2014||Publisher:||The Hong Kong Polytechnic University||Abstract:||Coherence of radar echoes is a fundamental observable in interferometric SAR (InSAR) measurements. It provides a quantitative measure of the scattering properties of imaged surfaces and therefore is widely used to study the physical processes of the earth. However, the estimated coherence is often biased due to the radar signal non-stationarity and the bias in the estimators used. Great efforts have been made over the past two decades to mitigate the errors in coherence estimation. Radar signal non-stationarity has been dealt with either by compensating for the systematic interferometric phase in the estimation window or by selecting and using the homogeneous pixels to avoid the texture effect in SAR images. The bias of the estimators has been corrected by the probability model deduced under Gaussian scene. Although the existing studies have improved the accuracy of coherence estimation with different levels of success, some key problems still remain. For example, it is difficult to avoid the overestimation of the coherence over noise only areas due to the overcorrection of the fringe pattern if no fringe pattern exists. It is also unclear how to mitigate the bias of the sample coherence when the sample size is small. In addition to the technical limitations, the assumptions behind these methods, such as the Gaussian property and the independence between the neighboring sample coherence, are often too rigorous to hold over many natural scenes, leading to mis-estimation of the coherence in the real world. The purpose of this thesis is to gain a better understanding of the sources of errors in InSAR coherence estimation, and to develop self-adaptive algorithms with fewer assumptions to solve the problems aforementioned.
We begin by briefly reviewing the existing techniques for coherence estimation. Three principal errors (i.e., errors due to biased estimators, appearances of image textures and fringe rates in estimate windows) are quantitatively analyzed by means of mathematic descriptions. Under the framework of multi-temporal InSAR (MT-InSAR) with moderate to large stack size, we propose a hybrid processing chain to mitigate three types of errors. To avoid overestimation of coherence induced by image texture, an adaptive two-sample distribution-free test is developed to compare the statistical homogeneity between two spatial pixels by using their temporal samples. To avoid underestimation or overestimation of coherence induced by local fringe rates, we suggest using the phase standard deviation map to guide the Fourier kernel adaptively. A newly developed estimator of bias correction, namely double bootstrapping, is deduced under assumption-free condition. The method is especially effective for small sample problem in which the biased coherence cannot be corrected by the existing estimators. Based on the foregoing processing chain, further progress has been successfully made for small SAR stacks. Statistically homogeneous neighbors for each central pixel are selected by using their spatio-temporal samples, rather than temporal samples only. Furthermore, considering the computational complexity of double bootstrapping, a Jackknife-based method is proposed for bias mitigation in coherence estimation. We present experimental results with both simulated and real data sets, and compare the performance of the proposed approaches against some of the existing ones. The results demonstrate that the new approaches can suppress the errors more effectively under various circumstances. Finally, by associating a decorrelation model with the new processing chain, we decompose coherence observations and extract the temporal components of decorrelation from a texture-significant area in Macau, and find that the time series of coherence are less noisy and biased than those obtained from conventional methods in almost all land covers. The results confirm that the methods presented in the thesis can improve the accuracy of InSAR coherence-based applications to earth observations.
|Description:||x, 112 pages : illustrations (some color) ; 30 cm
PolyU Library Call No.: [THS] LG51 .H577P LSGI 2014 Jiang
|URI:||http://hdl.handle.net/10397/7381||Rights:||All rights reserved.|
|Appears in Collections:||Thesis|
Show full item record
Files in This Item:
|b27628516_link.htm||For PolyU Users||203 B||HTML||View/Open|
|b27628516_ir.pdf||For All Users (Non-printable)||4.7 MB||Adobe PDF||View/Open|
Citations as of Jun 18, 2018
Citations as of Jun 18, 2018
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.