Phase Unwrapping Via Graph Cuts:in InSAR Applications

Interferometry Synthetic Aperture Radar (InSAR) technique is a developing and potential microwave remote sensing technology which has been more than30years since its inception. The initial use is to obtain the Digital Elevation Model (DEM) and mapping. It has gradually extended to differential interferometry techniques (D-InSAR) and has been applied to the measurement of small changes in topography. InSAR technique has showed great prospects on military, target tracking, earthquake deformation, volcanic activity, glacial drift, urban subsidence and landslides, etc.Interferometry of InSAR uses two images acquired from different location at same time or different time with a little position difference to form interferogram. Then DEM terrain can be calculated out through data processing.Phase unwrapping is the most important and difficult step in InSAR data processing and is also the major source of deviations in DEM. Noise and discontinuous phases are two common phenomenons in phase data; itthey would produce seriously error in some cases. They are two major problems in phase unwrapping algorithms. Residue is often introduced to describe phase problem caused by noise and discontinuous phases. It is the key factor to be considered in error problems. Algorithms currently being used are also mostly discussing aboutdistinguished by the processing of residues. Traditional PU algorithms can be roughly divided into three categories:(1) Identify the optimized integration pathpath of integration to avoid damage caused by error the wrong global propagation handicap of phase-residual;(2) Basing on the wrap phase to estimate and calculate unwrapping phase gradient, deviationserrors caused by residues will be adjust by Sqaure adjustment mean processing;(3) Do nothing for deviationserrors caused by residues, it will be retained in unwrapping results. The first category of unwrapping algorithms is commonly usually known as path integral method, the common methods including:Goldstein branch-cut. Quality guided path following method etc:The second category of unwrapping algorithm mainly refers to the minimum norm method such as Least Squares; The third category contains unwrapping algorithms like network planning algorithm.In recently years, there has developed many new algorithms such as image classification method, weighted iterative greedy method, Kalman filtering method, ant colony algorithm and graph cutsgraphy-partition methods etc. Phase unwrapping via graph cuts is the most efficiency and representatively method according to the merely abroad reference documentation. The two teams of France and Portugal have just started to make the research it. It seems that this algorithm has few teams to achieve whether on moulding or just calculation because of its difficulty and complexity on the basis of their ducumention. However, they also have acquired excellent results and received wide spread attention from the European Space Agency and other well-known universities or research institutions.Currently, papers about phase unwrapping via Graph cutsgraph cuts are very poorly limited in the international internationally and there is no paper can be found in China. Therefore, making deep research on phase unwrapping algorithm which on the basis of graph cuts does play a significant role in InSAR, not only on theory but application included. It has a very high significance in theory and application to be researched.The key problem of this method is to take maxflow/mincut to optimize the minimum energy of Markov Random Fields combining with Bayesian criterion. As we all know, this kind of energy optimization is a difficult task in Markov random field problem solving. Compared with other optimization algorithms, graph cuts are more complicatedcomplexity, difficulty and challenging. Graph cuts were used in computer vision introduced to MRF energy optimization area at first was considered to the problem of Low level computer vision originally. Teams of Boykov and Kolmogorov have published many papers about this algorithm. But it is nevertheless, it’s the teams of Dias and Tupin which introduced graph cuts to phase unwrapping algorithms for the first time. They have published some papers since2001. They have published related treatises from2001till now.In terms of its complexity and excellent performance on phase unwrapping algorithms, research and development of this algorithm has been seen as the goal of this master’s degree thesis. To focuseing on the research expounded from its theoreticaland practical sides, I need to gather a wide variety of information collect paper data, carry out the theoretical analysis theoretical analysis, and theoretical modeling, with the procedure of optimization and programming etc. Both of theoretical and experimental methods have been taken. Towards the whole process of research, graph cuts are also called maxflow/mincut. The computing of maxflow/mincut is the key and difficult step of these algorithms. In this thesis. Markov Random Fields theory. Bayesian criterion, energy minimization algorithm, graph theory and optimization theory are involved in phase unwrapping algorithm. All of these algorithms have played an important role in it. The way of Boykov-Kolmogorov augment path algorithm gives me the ideas to programming, meanwhile, its algorithm framework is adopted in this paper. But code part which has contained substantial of my new ideas and creation is definitely, it is not equivalent to Boykov-Kolmogorov’s. Programming about phase unwrapping based on graph cuts has completedly finished.finally. It has shown a great advantage over traditional algorithms of phase unwrapping. This algorithm has independent intellectual property right (more openchannels of information can not be found in the published literature).To verify the reliability, flexibility, capability of algorithms, this thesis designs some artificial phase data to confirm at experimental aspect. Especially some manual datasets on noises and discontinuous phases are designed to be verified emphasizly, and all of them have obtained good performance. Otherwise, in order to verify the correction of artificial simulating phase data, Radarsat-2phase data is used at the end of the experimental part of this paper. The interferogram of Radarsat-2is processed by Nest software which designed by European Space Agency (ESA). All of these results show the huge superiority than traditional methods. It also proves the effectiveness of phase unwrapping via graph cuts. To verify algorithms,64groups of experiments have been completed. It has obtained a number of valuable results and conclusions on the foundation of analysis.Random error evaluation is also an important aspect to judge algorithms, rewrap mean square errors (RMSE) is an indicator of performance. Time-consuming is also taken into consideration. Though experimental data have been simulated, there also have errors of histogram. The study of this paper is not only instructive for the study, development and application such as Synthetic Aperture Radar, Magnetic Resonance Images (MRI) and optical objects indicate of profile measurement, but also a great significance for the development of related software platform and applications in China.