Investigation On Insar Phase Unwrapping Algorithms And Software Development

Synthetic Aperture Radar Interferometry (InSAR) and Differential InSAR (D-InSAR) are microwave remote sensing technologies developed since the1960s. Compared with the traditional remote sensing technology, InSAR has many technological advantages, such as large coverage, working at all time and under all-weather conditions. It has been widely used in the extraction of digital elevation model (DEM) and the surface deformation measurements.Phase unwrapping is one of the key steps and technical difficulties in the data processing of InSAR and D-InSAR. Its accuracy will directly affect the precision of the DEM extracted by InSAR and the ground deformation measured by D-InSAR. In the past30years, scholars at home and abroad have proposed many representative algorithms of phase unwrapping. According to the strategies of unwrapping procedures, they can be divided into two categories:the path-following method and the minimum norm method. In specific applications, phase unwrapping is mainly designed as an embedded module included in the InSAR and D-InSAR data processing softwares, and there is no such professional software which is specifically developed for phase unwrapping. Moreover the current softwares only cover around one to two kinds of unwrapping methods, which results in poor selectivity and adaptability. Furthermore, current researches on phase unwrapping often focus on improving the unwrapping algorithms, and there is lack of comprehensive comparation analysis and evaluation of different algorithms.This article emphasis on stating the processing procedures and implementation methods of four path-following methods (Goldstein branch cut algorithm, quality-guided path-following algorithm, mask cut algorithm and Flynn’s minimum discontinuity algorithm) and four minimum norm methods (discrete cosine transform algorithm, unweighted multigrid algorithm, preconditioned conjugate gradient algorithm and minimum Lp-norm algorithm) based on introduction of the basic principles of phase unwrapping. On the Windows operating system, combining with the object-oriented C#language and ArcEngine which is the embedded component library of GIS, a InSAR phase unwrapping system which is dedicated to phase unwrapping was developed. To improve the success rate and reliability of phase unwrapping, the man-machine interactive operation and the evaluation strategy were introduced in the process of phase unwrapping, i.e., add modules to the system which include unwrapping evaluation module, DEM generation module, module of manually modify the unwrapping path (i.e., the branch lines) and box selected unwrapping module. In addition, to improve the effect of unwrapping, the system also provides the sub-module of image processing.Except GAMMA software, the InSAR phase unwrapping system defines the same data structure as other InSAR and D-InSAR processing softwares which are commonly used, and thus it allows data sharing with other softwares. What’s more, the system provides eight kinds of phase unwrapping methods, so users can select the appropriate unwrapping method according to different types of data. With the aid of the unwrapping auxiliary modules to improve the unwrapping results, the precision of the eventually acquired DEM and ground deformations can be promoted.Finally, both the simulated wrapped phase data and the interferometric phase data of southeast Lincoln county of United States were used to test the execution efficiency and stability of the InSAR phase unwrapping system, and to evaluate the accuracy of the eight algorithms in the system. The characteristics, advantages and disadvantages of these algorithms are as follows:First, least-squares algorithm based on the discrete cosine transform (DCT) required the least amount of memory and with the fastest execution speed. But there is higher probability of resulting in inconsistencies between the results and the initial phase and the results often appears smoother, which is similar with the unweighted multigrid algorithm. Therefore, unweighted least-squares method has poor applicability in practice, so it is generally not recommended to be used.Second, Goldstein branch cut algorithm has high execution speed and low memory needs. The precision of unwrapping result is higher when the phase has high signal-to-noise ratio, so it is selected in the first place in the practical application. But for the areas where the quality of interference maps vary remarkably, the results of unwrapping are prone to be discontinuous or many isolated areas may be induced. However, after using of the manually modified branch-cut lines and box selected re-unwrapping operation which provided by the system, the effect of unwrapping can be improved, they made the standard deviation of the unwrapping phase of the "mountains" in the experiment decreases from0.099rad to0.047rad and the standard deviation of the corresponding DEM decreases from3.409m to1.698m. The accuracy has been greatly improved, and the effect can be comparable with the minimum Lp norm algorithm, but its operation speed is much faster than the minimum Lp norm algorithm.Third, when good quality map is available, we can choose quality-guided path-following algorithm, mask cut algorithm and preconditioning conjugate gradient algorithm. If these algorithms are all failure, then the Flynn’s minimum discontinuity algorithm and minimum Lp-Norm algorithm can be considered. Especially, the mean of differences between the unwrapped phase from minimum Lp-Norm algorithm and minimum cost flow algorithm in SARscape is3.119rad, and the standard deviation is only0.041rad, the mean of the differences between the corresponding DEM is0.087m and the standard deviation is1.439m. However, Flynn’s minimum discontinuity algorithm and minimum Lp-Norm algorithm consume more memories, and the execution speed is slower with minimum Lp-Norm algorithm is the slowest one.Forth, in general, the unwrapping results of least-squares methods are smoother, it can unwrap the phase in the noisy areas, but at the same time, the errors will propagate to the high quality areas, resulting in systemic bias. Path-following methods will isolate the noise areas and the regions where the change of phase gradient is large and it limits the phase in such areas to participate in the unwrapping, thus can avoid the transmission of errors. As a result, in areas where the signal-to-noise ratio is high and gradient change is small, we can choose Path-following methods, while in areas with low signal-to-noise ratio or high phase gradient change rate the least-squares methods seem more appropriate.