Research On Phase Reconstruction Method Based On Probability Generative Model For SAR Image Stacks

Surface deformation may cause a series of serious disasters,such as ground fissures,sea water flooding,landslides and so on.These geological disasters could threaten the safety of human life and properties.Monitoring surface deformation is helpful for people to study their spatial and temporal distribution and formation mechanism.In recent years,multi temporal interferometric synthetic aperture radar has shown great potential and irreplaceable.However,its performance is easily affected by the decorrelation,which makes it difficult to interpret the pattern of surface deformation comprehensively.The phase reconstruction technology arises to reduce the influence of decorrelations.There are two main approaches to reconstruct interferometric phases,which are based on maximum likelihood estimation and eigen-decomposition,respectively.The eigen-decomposition method has been proved to have high computational efficiency,and it is considered to be the best choice for processing large time dimension synthetic aperture radar image stacks.The satellite revisit period is gradually shortened,which means the image stacks are increasing in dimension.Even the traditional eigen-decomposition method could not guarantee the completion of phase reconstruction in a reasonable time.Aiming at the computational efficiency,this thesis proposes a new method to reconstruct interferometric phase.The basic idea of this method is to build a probabilistic principal component model for distributed scattering signals,and transform the estimation of interferometric phase into the solution of model parameters.The experimental results show that the phase reconstruction method based on probabilistic principal component analysis has great computational efficiency advantage in stack analysis for large dimension images.The main work and contributions of this thesis are as follows:(1)Deductive reasoning of phase estimation methods based on maximum likelihood and eigen-decomposition were completed,the essence of phase reconstruction is hence revealed mathematically.In the process of high-dimensional synthetic aperture radar image stacks processing,the traditional EVD method still has limitations.(2)According to the establishment and solution method of the real probabilistic principal component analysis model,a corresponding probability model is constructed for distributed scattered complex signals.By analyzing the mathematical form of model parameters' solution and following the idea of eigen-decomposition phase estimation,a phase estimation scheme based on probabilistic principal component analysis and the corresponding posterior quality evaluation index were designed.(3)A method for solving model parameters incorporating the maximum expectation algorithm is derived.In order to solve the problem efficiently,according to the phase reconstruction mechanism,an algorithm for solving the model parameters was designed and realized.Probabilistic principal component analysis method is proved to be effective and efficient.(4)The phase reconstruction method based on probabilistic principal component analysis is applied to real data.The synthetic aperture radar(SAR)images of Yunyang County and its surrounding areas in Chongqing are captured by sentinel-1A satellite.The phase reconstruction methods based on probabilistic principal component analysis and standard principal component analysis are compared and verified.It is further proved that the probabilistic principal component analysis method can greatly improve the efficiency while maintaining the estimation accuracy.