Radar Feature Enhancement Imaging And Altimetry Based On Bayesian Learning

Synthetic aperture technology can form a larger virtual antenna with a smaller real antenna,so as to achieve the high-resolution detection effect of a real large antenna.This technology is first applied in the radar imaging,namely,synthetic aperture radar(SAR)imaging.SAR is an all-weather,all-time imaging radar,widely used in military and earth remote sensing detection.Then,the synthetic aperture technology is applied to radar altimeter,and delay Doppler altimeter(DDA)is proposed.Compared with the traditional radar altimeter,DDA has the characteristics of higher accuracy.However,in SAR and DDA,the imaging and altimetry processing algorithms largely affect the detection performance of redar systems.In terms of SAR imaging,conventional imaging algorithm mainly rely on discrete Fourier transform processing,and the resolution is limited by the bandwidth.The development of compressed sensing(CS)technology has broken the limitation of Nyquist sampling theorem,so that super-resolution SAR imaging can be realized.To this end,the Bayesian learning super-resolution feature enhancement imaging method based on CS technology has become a current research hotspot.However,the current Bayesian learning methods have the following two problems: 1)The statistical prior is ususlly fixed,which leads to overfitting of imaging results.2)The imaging prior is single and cannot represent the rich imaging target features.In terms of altimetry,the traditional altimetry estimation algorithm such as least square(LS)estimation,does not consider the prior characteristics of the altimetry echo,resulting in limited accuracy.Therefore,in ordaer to solve the problems of fixed prior and single prior in Bayesian learning SAR imaging,the super-resolution feature enhancement imaging methods based on flexible prior and hybrid prior are proposed in this thesis,respectively.In terms of the problems existing in DDA altimetry,an altimetry estimation algorithm based on Bayesian learning is proposed.The specific research contents include the following three points:1.For the problem of the priori is usually fixed and lack flexibility in the Bayesian learning of SAR imaging,a super-resolution SAR imaging algorithm based on flexible priori Bayesian learning(FP-Bayes)is proposed.In this algorithm,the priori is designed as a flexible generalized Gaussian distribution(GGD),and the posterior of the imaging scene is derived through a Bayesian hierarchical framework.Considering that the posterior is complicated and may be nonsmooth,the proximal operator is to conbined with Hamiltonian Monte Carlo(HMC),and the proximal HMC(PHMC)algorithm is developed,which is expected to achieve the auto-learning sampling process.To this end,the high-resolution SAR imaging results is obtained.In the experimental part,the simulation and raw data imagery results have a high imaging resolution,which prove the effectiveness of the proposed FP-Bayes imaging algorithm.2.The problem of single prior in conventional Bayesian SAR imaging,which cannot represent rich and useful target features.Therefore,a decomposition-coordination Bayes(DCBayes)SAR imaging algorithm based on a hybrid prior is proposed.In this algorithm,the hybrid prior is composed of the Laplace distribution and exponential total variation(TV)distribution wich can characterize rich and useful imaging scene features.However,the intended hybrid prior lead to a complicated posterior,which cannot be solved by the conventional MCMC algorithm.Therefore,a DC-Bayes solution framework is proposed.In this framework,the hybrid prior imagery problem can be decomposed into two local single prior imagery sub-problem and solved separately in the decomposition step,the coordination step reconciles the local sub-problems by introducing dual variables,which is expected to ahieve efficient solution of the global imaging models.Finally,Both simulation and raw SAR data experiments demonstrate the effictiveness of the DC-Bayes imaging algorithm.3.In terms of the problem of the limitation of the DDA altimetry accuracy,a newly Bayesian learning altimetry algorithm based on the MCMC sampling is proposed in this thesis.In this algorithm,the altimetry parameter is modeled as Laplace distribution.Under the assumption of Gaussian likelihood,the posterior is deduced through the Bayesian hierarchical model.To this end,the MCMC sampling methods,such as HMC,are introduced to estimate the altimetry parameters.The result of simulation and raw DDA echo data experiments show that the proposed MCMC altimetry algorithm has higher accuracy than the traditional altimetry algorithm,which verify the superiorty of the proposed altimetry estimation algorithm.