Study On Key Technology Of Repeat-Pass Low-Frequency UWB InSAR

Cross-track synthetic aperture radar interferometry(In SAR)technique has been employed to construct digital elevation models(DEMs)of terrain surface with high spatial resolution and accuracy.As the SAR systems provide their own illumination,they can acquire information almost independently of meteorological conditions and sun illumination.This technique has been widely applied in the field of surveying and mapping.The low-frequency ultra-wideband In SAR(UWB In SAR)has the capability of foliage penetration and thus can perform topographic mapping of the ground beneath the foliage.Comparing with the normal In SAR system,the low-frequency UWB In SAR systems always work in the repeat-pass mode,and encounter some different problems in data processing.In this dissertation,some key technologies for the data processing of the low-frequency UWB In SAR systems are studied,and the major work is focused on image registration,phase unwrapping(PU)and absolute phase determination.Based on the work of the formers,some new methods are proposed to improve the effiency of data processing for the low-frequency UWB In SAR.The main content of this dissertation is summarized as follows:1.The image registration for the In SAR data is studied.By analyzing the geometric difference between the two In SAR images,the warp function between the master image and the slave image is discussed,and theoretic analysis shows that the magnitude of the mutual shifts between the two images is proportional to the topography and the fractional bandwidth of the system.For the low-frequency UWB In SAR system with fractional bandwidth larger than 0.2,the warp function cannot be fitted by the low-order polynomials.It has to estimate the shift for each pixel.Based on the above conclusion,a basic image registration algorithm is proposed in this dissertation,which estimates the shifts by cross-correlation and interpolation operations.Besides,we refer to the reference phase derived from the registration offsets as the registration phase(RP),and the subtraction of RP from the interferometric phase as the misregistration phase(MRP)which is proportional to the registration error.Therefore,we propose to evaluate the accuracy of registration by calculating the errors from MRP which is more intuitive and accurate.2.The PU method based on graph theory is studied.The L1-norm based PU can be solved by the minimum cost network flow algorithm(MCF),which is a classic graph theory algorhitm and can be implemented easily by using the standard algorithm.MCF can provide the high quality PU result,however with the very heavycomputional burden.To sovle this probem,the network flow model associated with PU problem and the characteristic of the branch cuts in the optimal solution is analyzed.It is found that there are only dipole cuts in the optimal solution.Therefore,the minimum cost network flow problem can be simplified to a minimum weight perfect matching(MWPM)problem on a weighted bigraph associated with PU problem,and the proposed method based on this theory is called MWPM algorithm for short.Comparing with MCF,MWPM has the lower computing complexity and the same unwrapping quality.3.The fast PU method based on minimum discontinuity is studied.Since the graph theory algorithm has the strongly polynomial complexity,it doesn't apply to the phase image which contains many residues.In that case,we choose Flynn's minimum discontinuity algorithm to solve the PU problem,which is also a L1-norm based PU algorithm.Flynn's algorithm also has the problem of high computing complexity.However,it is found that the computing complexity of the algorithm depends on the size of the image and the number of the discontinuties within it.Therefore,we propose a pre-unwrapping assisted minimum discontinuity(PAMD)PU method,which has high efficiency and the same unwrapping quality as Flynn's algorithm.In the PAMD method,the phase image is firstly unwrapped by the pre-unwrapping algorithm which generates and optimizes the dipole cuts.The pre-unwrapping algorithm is quite fast and the obtained result is much close to the optimal solution of minimum discontinuity,therefore it can be further optimized by Flynn's algorithm efficiently.Besides,for the low-frequency UWB In SAR data,we proves that the MRP image can be unwrapped by Flynn's algoritm directly with very small computing complexity due to that the MRPs in the area with the high coherence are always unambiguous.Therefore,the unwrapped interferometric phase can be obtained by unwrapping the MRP image with high efficiency.4.The phase unwraping method for the large-scale interferograms acquired form the low-frequency UWB In SAR data is studied.A region-partition-based PU method is proposed,which could improve the efficiency and decrease the processing memory of the large-scale PU problem.The MRP is unambiguous in most of the area with high coherence because of the large fractional bandwidth;thus,the regions that assuredly have the unambiguous MRP are partitioned out from the MRP image and referred to as the high-quality area.Meanwhile,the remaining low-quality areas are separated by the high-quality areas and are considered irregular tiles.After unwrapping the tiles by a local PU algorithm either in parallel or in series,the full-size unwrapped result is obtained.The test performed on real P-band UWB In SAR data confirms the effectiveness and efficiency of the proposed method.5.The residual phase ambiguity estimation for the unwrapped phase image is studied.The traditional split-spectrum algorithm and the residual delay estimation method(RDE)are difficult to be implemented or have the high computional complexity.In this dissertation,a new estimation method for residual phase ambiguity is proposed,which is based on the distribution of the unwrapped MRPs in the area with high coherence.The proposed method has the advantage of easy to be implemented.Furthermore,the theoretic analysis and the experimental results show that the proposed method has the same accuracy as RDE and has the much lower computional complexity than RDE.