Research On Two-Stage Programming Approach Framework And Phase Unwrapping For Multibaseline InSAR

Interferometric Synthetic Aperture Radar(InS AR)is a well-developed remote sensing tech-nology,which can accurately measure the important physical parameters of earth,for exam-ple,the topographic mapping,the surface deformation of the earth and the movement of the glacier.Since the measurement obtained by radar is the principal value of absolute phase,phase unwrapping(PU)has become an indispensable key step in InS AR technology,and its solving performance directly determines the quality of the subsequent remote sensing prod-ucts.In the past few decades,the traditional single baseline PU(SB PU)and multibaseline PU(MB PU)are almost developed independently and each has its own advantages.The SB PU is well developed and has good noise noise robustness and high efficiency.However,due to the constriction of the phase continuity assumption(that is,the observed area should have good spatial continuity),it can hardly solve the complex terrain(e.g.valley and cliff)cor-rectly.Compared with the SB PU,the MB PU can break through the limitation of the phase continuity assumption,but it is difficult to meet the needs of practical applications due to its poor noise robustness.In recently years,a novel idea of combining SB PU and MB PU has been proposed and some preliminary research results indicating that this idea can effectively improve the noise robustness of MB PU methods.Continuing with this idea,there are a series of problems need to be studied.Firstly,the classical SB PU methods have good noise robustness,and how to transplant them to MB PU domain needs to be studied.Secondly,it is necessary to establish a general framework for solving MB PU in order to construct more new MB PU methods with good noise robustness;Besides,the fast-developing InSAR technology also faces a new challenge.As the scale of observation data becomes larger,it is necessary to find an appropriate compromise between computing speed,calculation re-source consumption,and calculation accuracy.Aimed at these problems,the main work of this thesis is summarized as follows:1.A multibaseline InSAR two-stage programming approach(TSPA)processing framework is proposed,which is referred to as the TSPA framework for short.The PU methods under the TSPA framework are collectively referred to as the TSPA-PU methods.In the Stage-1 of TSPA framework,it estimates the phase gradient information of each interferogram by combining multiple interferograms according to CRT,and afterwards it applies the es-timate to solving the absolute phase in the Stage-2.The significance of the multibaseline TSPA framework is that it allows the optimization models of the classic PU methods to be transplanted into the Stage-2 of the TSPA framework.For example,the optimization models of SB Lp-norm(p=1,2,?)and the Minimum infinity-norm-based method(ab-breviated MIN method)can be easily transplanted into the TSPA framework.In addition,more forms of models can be introduced into the Stage-2,such as outlier-detection-based PU iterated method(OD method)and branch-cut method(BC method)using residues to generate branch-cut.By transplanting the above methods into the TSPA framework,TSPA-Lp(p=2,?),TSPA-MIN,TSPA-BC and TSPA-OD PU methods are proposed in this thesis.The results of the experiments indicating that TSPA framework can greatly improve the PU accuracy of SB PU methods and improve the processing performance of complex terrain to a new level.TSPA framework also introduces the important concepts"residue"and "branch-cut" in the MB PU domain.Therefore,TSPA framework brings more research way to the study of MB PU.Those problems that have been encountered in the development of SB PU will be expanded in the MB InSAR field.2.In the traditional SB PU methods,the estimation of the gradient is based on the assump-tion of phase continuity,which causes the estimated value of the gradient to be limited within 2?,whereas for MB PU,it applies Chinese remainder theorem(CRT)to estimate the phase gradient,which ensures that the absolute value of the gradient estimate can exceed the limit of 27r.However,the sensitivity of CRT to noise makes the robustness of gradient estimation poor,which leads to unsatisfactory PU results.For the Stage-1 of TSPA framework,a new method of absolute phase gradient estimation was proposed in this thesis by applying the local plane model(LPM),in which the local area of the interferogram is regarded as a plane and then the phase gradient is estimated by combining all the pixels in the local plane.From the perspective of signal processing,since the terrain gradient contained in each interfero-gram is correlated and the noise are considered to be independent,by combining pixels of multiple interferograms in the local plane,the influence of noise on CRT can be suppressed and the estimated results are more reasonable.The effective improvement of the accuracy of the gradient estimate can significantly improve the PU accuracy of the TSPA-PU methods.Both theoretical analysis and experimental results validate the effectiveness of LPM-TSPA method.3.With the increasing data scale of the single interferogram of InSAR,the big data pro-cessing has become a challenge to SB PU and it is more challenging for MB PU due to the multiple interferograms need to be processed.This thesis adopts the strategy of " divide and conquer" to process the big data,and proposed a large-scale MB PU based on the convex hull and cluster analysis under the TSPA framework(abbreviated as CCFLS method).In the Stage-1 of CCFLS,estimating the gradients and computing the residues of MB and in the Stage-2,clustering the residues and generating the convex hull at the same time.When the clustering is finished,the convex hull of every residue cluster,which can be used to replace branch-cut,is also obtained.The PU results of large-scale InS AR data can be obtained by a simple path integration that bypass convex hull.Because the speed of generating convex hull is much faster than that of branch-cut,the efficiency of CCFLS is greatly improved.It is a reasonable to believe that this method makes a reasonable compromise between the solving speed and the area to be solved,because the area surrounded by the convex hulls is usually a low-quality area in the interferogram.For applications with high real-time requirements,it is worthwhile to abandon the meaningless processing of low-quality area in exchange for the improvement of processing efficiency.Experiment verify that the effectiveness of the CCFLS method is significantly improved compared to the original TSPA method.