Study On The Cubature Kalman Filtering Based Phase Unwrapping Methods

Interferometric synthetic aperture radar(In SAR) can all-weather and all-time obtain efficiently digital elevation model(DEM) of the target area, it has been widely used in terrain mapping, surface deformation monitoring, target detection, and other fields. As a key step of the In SAR technology, the results of phase unwrapping success is directly related to the extraction of the correctness and accuracy of target elevation information, and it has been a hot topic of In SAR technology research. This dissertation focuses on the key issues of Kalman filtering based phase unwrapping which can be used in processing the interferograms featured with intensive fringes and high-level noise level. The contents of the dissertation include non-linear model of the Kalman filtering based phase unwrapping, the effect of multi-look strategy and pre-filtering for Phase Unwrapping, the effect of quality map for Phase Unwrapping results, the phase unwrapping strategy for interferograms with uncontinuous quality map. The main results are as follows:(1)Traditional phase unwrapping methods which require pre-filtering can achieve desired unwrapping results for the interferograms with simple fringe interferogram. For the raw unfiltered complex interferogram with intensive fringes, these phase unwrapping methods could not successfully unwrap the interferometric phase. Phase unwrapping method based on Kalman filtering was studied thoroughly. Because state equation of phase unwrapping model is linear while the observation equation are non-linear, we proposed a simplified CKF phase unwrapping algorithm, which can reduce the computation burden of the state prediction step. Phase gradient estimation is one of the key steps of Kalman filtering based phase unwrapping,which will affect the phase unwrapping results directly. Some factors of phase gradient estimation such as signal noise ratio(SNR), window size and sampling number were studied. Simulated and true interferograms tests both show that Kalman filtering based phase unwrapping method is more suitable for the interferograms with high-density and high level noise. Then, the effect of nonlinear filtering model was analyzed. Both the theoretical analysis and experimental results show that EKF phase unwrapping model and CKF phase unwrapping model can achieve almost the same results for simulated interferograms while CKF performs better for real interferograms of mining area.(2)The effects of multi-look processing and pre-filtering for Kalman filtering based phase unwrapping were studied. The selection rules and strategy of multi-look processing and pre-filtering was proposed. The experiment results show that for the interferograms with high level noise and dense fringes, multi-look processing and pre-filtering can improve the signal noise ratio while lose stripe detail information and other features. Taking the denoising ferature of Kalman filtering itself into account, the number of looks can be as small as possible. Also, a slight pre-filtering step can both reserve the fringe details and remove speckle noise. Experiments show that, when using Kalman filtering based phase unwrapping, the pre-filtering strategy can select a minimum number of interferograms and a slight pre-filtering.(3)The effect of different quality map for Kalman filtering based phase unwrapping was studied. The six traditional six quality map index for guiding the phase unwrapping path were analyzed, compared. Both the theoretical analysis and experimental results show that the Fisher Distance index is more robust especially for the actual interferogrms with high noise and many complex types of covers because Fisher Distance index considers both the pixel phase peculiarity and spatial coherence information. However, the prime fisher distance index only considers the similarity degree between the current pixel and eight surrounding pixels while ignores the similarity information among the eight surrounding pixels. Based on this, a modified Fisher Distance index was proposed which takes the similar information among all the pixels in the window. Simulation and actual interferograms phase unwrapping results show that the modified index performs better.(4)A multi-reference points strategy for Kalman filtering based phase unwrapping to process the interferograms with discontinuous quality map was proposed. Compared with single reference point strategy, the proposed one can deal with discontinuos or isolated area and achieve better performance. Furthmore, a region growing method with multi-reference points was proposed. In applying this strategy, a selection method for reference points is presented, which overcomes the subjectivity and unpredictability in selecting reference points. Firstly, the interferogram is partitioned into high and low quality areas by applying a threshold to the quality map. Then, a proper seed is selected in a bigger isolated high quality area. A path map will be calculated according to the seed and the initial quality map. Both simulated and real data experiments results validate the feasibility and effectiveness of the proposed multi-reference point strategy and reference points selection method in overcoming error propagation.