Research On InSAR Phase Unwrapping Based On High Performance Computing And Development Of A Cloud Platform

High Performance Computing (HPC) generally refers to the practice of aggregating computing power in a way that delivers much higher performance than one could get out of a typical desktop computer or workstation in order to solve large problems in science, engineering, or business. In the field of Earth exploration, with rapid development of remote sensing technology like the interferograms of synthetic aperture radar interferometry (InSAR) are becoming larger and larger, HPC technology is becoming a new requirement of information extraction from remote sensing images.In the interferometric technique, very accurate measurements of the Earth’s topography or its surface deformation are derived from radar-image phase data. Phase, however, is defined only modulo 2π rad, so resulting two-dimensional array of measurements is wrapped into the principal interval of (-π,π) radians with respect to some modulos or ambiguity, producing the wrapped phase. The processing of estimating the true phase form the wrapped phase is called phase unwrapping. Because of the presence of the noise, undersampling and object discontinuities, phase unwrapping is intractable and nontrival. So far, many phase unwrapping algorithms have been proposed. These algorithms are advanced in different aspects, such as the accuracy of the solution or the speed. However, the limitation of computer memory size is ignored in the design of most of these algorithms. With rapid development of the InSAR technology, the interferograms are becoming larger and larger. When the size of the interferogram exceeds the limitation of computing capability, adopting divide and conquer policy during the phase unwrapping is unavoidable in practice. Under this condition, whether the phase unwrapping result of each tile is consistent with that of the whole image or note becomes a new challenge for phase unwrapping.In this dissertation, a novel simulated annealing based phase unwrapping algorithm is proposed. In this algorithm, a random initial configuration is generated by the nearest-neighbor strategy, where all residues are connected by branch cuts in clumps. And then simulated annealing idea is used to further optimize the set of branch cuts. Finally, flood-fill method was used to unwrap phases avoiding these branch cuts. The ideal speedup of existing parallelized simulation annealing algorithms is the reason that the method of simulated annealing is employed to do combinatorial optimization for branch cuts. The algorithm is implemented using a hybrid parallel programming model that mixes a shared memory multi-threading model with a distributed memory multi-processing model, which results in a good speedup over the serial implementation. Using real and simulated interferometric data, we demonstrate that our algorithm is highly competitive with other common alogrithms in speed and accuracy. We also demonstrate that the proposed algorithm can be efficiently parallelized and performed across nodes in an HPC cluster. In addition, a novel tiling strategy based on the hybrid shared-distributed memory model is proposed. The proposed tiling strategy based on the nature of parallel computing guarantees the globality of phase unwrapping and avoids large-scale errors introduced. Using an HPC cluster with a lot of processors, the required memory in each processor is significantly reduced. We demonstrate that the proposed algorithm with this new tiling strategy successfully unwrap very large simulated SAR interferometric data sets without any tiling artifact.With the interferograms are becoming larger and larger, there is one more new challenge that big data of interferograms could not be processed and analyzed on a single computer. Therefore, Spark, a fast and general cluster computing system for big data, is used to data pre-processing like clustering for phase unwrapping. An execution platform is built combining with Spark, OpenMP and MPI, the latter two of which are used to run phase unwrapping algorithms. The execution platform is extended to a cloud computing platform based on HPC system. Moreover, it is integrated with existing computational mineral physics applications along with a new component developed based on Software as a Service (SaaS). It allows the user to access scientific software and hardware resources on-demand from anywhere in the world with a web browser, omitting complicated, multi-stated job submissions in original HPC environments and removing the need to use a Unix/Linux environment. It provides students of different backgrounds with an overview of numerical simulations method to study mineral physics, makes it easier to analyze output data of simulations and to manage useful results efficiently, and reduces human error.