Study On High-efficiency And High-precision Filtering Methods For Synthetic Aperture Radar Interferometric Phase Images

Synthetic Aperture Radar Interferometry(InSAR),being as a means of microwave remote sensing which is both active and penetrative,enables high-precision measurement of geophysical parameters of the Earth such as Digital Elevation Model(DEM),ground deformation and subsidence as well as glacier movements.The interferometric phase is the most important physical quantity in the whole procedure of InSAR processing,the accuracy of acquired geophysical parameters of which greatly depends on the phase quality.However,affected by various decorrelation factors,the interferometric phase is always contaminated by much noise which severely affects the follow-up processing steps.Therefore,the interferometric phase has to be filtered before further processing.While the classic phase filtering methods are very efficient,there is still a lot of room for improvement with respect to filtering accuracy.The filtering methods proposed recently are much better than the classic ones in terms of filtering accuracy,but their efficiency is very low.As a result,filtering accuracy and efficiency are not simultaneously considered in existing filtering methods.In this dissertation,the interferometric phase filtering problem is studied and some high-efficiency and high-precision filtering methods for interferometric phase are proposed.The main content of this dissertation is introduced as follows:In Chapter 3,the unique characteristics of interferometric phase filtering such as phase noise model,types of noise and purpose of filtering are summarized.Then,the main and important interferometric phase filtering methods are classified into three groups: spatial-domain methods,transform-domain methods and methods based on split-and-conquer strategy.The basic principles and steps of these methods are introduced in detail,together with their essence and drawbacks.The types and principles of iterative filtering are given,and the fact that,as a simple yet very effective means,the iterative filtering has been incorporated by many phase filtering methods has been pointed out.Third,the ways to evaluate phase filtering results are summarized and the quantitative indexes are presented in detail.Four new indexes: absolute filtering performance,relative filtering performance,noise suppression performance and detail preservation performance are presented,and quantification of these indexes are realized based on previous research from other researchers,making quantitative comparisons between different filtering results possible.Finally,the effect of some quantitative indexes and filtering performance of some phase filtering methods are shown through experiments carried on simulated data.In Chapter 4,the modified patch-based locally optimal Wiener interferometric phase filtering method—MPLOW is studied.First,the patch-based locally optimal Wiener filtering method(PLOW)is introduced and how the geometric similarity and photometric similarity are utilized to obtain high-precision results is explained in detail.Second,the reasons why the PLOW method is not suitable for interferometric phase filtering are analysed in detail and three corresponding improvements such as coherence-based image patch clustering,coherence-weighted estimator for cluster mean and locally adaptive mean estimator for noise covariance are made,forming the MPLOW method which is suitable for interferometric phase filtering.Finally,the effectiveness of MPLOW is verified through both simulated and real data sets.Experimental results indicate that the filtering accuracy of the MPLOW method is close to that of the SpInPHASE method,but the efficiency of the former is about 6.6 times that of the latter.In Chapter 5,the adaptive spatial-domain and frequency-domain interferometric phase filtering method—ASFIPF,which is derived from the Baran method and fast nonlocal mean method,is studied.First,three interferometric phase filtering methods based on the split-and-conquer strategy are reviewed and analysed,and the respective problems of these methods are given.On one hand,the approaches for extractions of the low-frequency components of the phase signal of these methods do not make full use of the different statistical characteristics of the signal and noise in the frequency domain,and these approaches are not robust to the variation of noise level.On the other hand,the approaches for extracting high-frequency components of the phase signal of these methods are not accurate enough.Second,the ASFIPF method is proposed according to the problems analysed above.In ASFIPF,the frequency-coordinate-based hard thresholding of the amplitude of the frequency spectrum and power operation of the amplitude of the frequency spectrum are used in the extraction of the low-frequency components,and the modified fast nonlocal mean method is used in the high-frequency components recovery.To obtain high-precision filtering results,some filtering parameters are related to the coherence and pseudo-correlation while others are experimentally determined through simulated data using the 2? and the minimum variance rules.Finally,the effectiveness of ASFIPF is verified using both simulated and real data sets.Experimental results indicate that the filtering accuracy of the ASFIPF method is on par or exceeding that of the SpInPHASE method under many situations.Furthermore,the efficiency of the ASFIPF method is about 93.7 times that of the SpInPHASE method.In Chapter 6,the adaptive and iterative phase filtering method in the frequency domain—AIPFFP,which is based on the second iterative filtering,is studied.First,the filtering framework that consists of the frequency-coordinate-based hard thresholding of the amplitude of the frequency spectrum and power operation of the amplitude of the frequency spectrum is determined based on exiting frequency-domain iterative filtering method,and the effectiveness and limitations of this framework are shown through one simulated data set.Second,for the purpose of high-efficiency and high-precision,the AIPFFP method is proposed.In AIPFFP,filtering is performed two times,and the same filtering framework with different sets of filtering parameters is used to obtain results with high-efficiency and high-precision.Some filtering parameters of the AIPFFP method are related to the coherence and pseudo-correlation while others are experimentally determined through simulated data using the 2? and the minimum variance rules.Finally,the effectiveness of the AIPFFP method is verified using both simulated and real data sets.Experimental results indicate that the filtering accuracy of the AIPFFP method is on par or exceeding that of the SpInPHASE method under many situations.Furthermore,the efficiency of the AIPFFP method is about 47.5 times that of the SpInPHASE method.