Research On Sparse Aperture Inverse Synthetic Aperture Radar Imaging Withing Bayesian Framework

With the ability of achieving high-resolution radar image of moving targets to acquire their size or structure information,the inverse synthetic aperture radar(ISAR)imaging technique is essential to feature extraction of radar targets,and has been put into various civil and military applications.For the sparse aperture data,however,the traditional range-Doppler(RD)ISAR imaging method would suffer from low resolution and high sidelobes.This thesis focuses on the challenge and difficulties of sparse aperture ISAR imaging techniques within the sparse Bayesian framework.Main works include researches of the sparse prior,sparse aperture ISAR autofocusing,sparse aperture ISAR cross-range scaling,sparse aperture Bistatic ISAR(Bi-ISAR)imaging and sparse aperture interferometric ISAR(InISAR)imaging.Firstly,the background and significance of this thesis are introduced,and the developments of the high resolution wideband radar system and ISAR imaging techniques are given in the introduction.Also,the framework of this thesis is presented.Then,during the theoretical research of the sparse Bayesian reconstruction,two sparse priors,including the log-Laplacian prior and the Laplacian scaling mixture(LSM)prior,are proposed.Further,the sparse Bayesian reconstruction based on these two sparse prior are derived,in which the maximum a posterior(MAP)method and the variational Beyesian method based on the Laplace approximation are utilized to recover the sparse signals that are modeled with the log-Laplacian and LSM prior,respectively.Different from the MAP method,which basically belongs to point estimation,the Laplace approximation based variational Bayesian method for LSM prior is the full Beyesian inference,and is able to obtain the posterior of the sparse signal as well as its high moments,which is potentially useful.Finally,experimental results based on both simulated and measured data validate the effectiveness of the proposed algorithms.For sparse aperture ISAR autofocusing,two algorithms based on the joint constraint of minimum entropy and sparsity are proposed.First,the LSM prior based signal model for sparse aperture ISAR autofocusing and azimuth compression is constructed.And then,the variational Bayesian method based on the Laplace approximation is utilized to reconstruct the ISAR image,and the phase error is estimated by minimzing the entropy of the ISAR image during the iteration.The images that are selected to estimate the phase error in the proposed two autofocusing methods are different,which are results of the RD imaging and the sparse Bayesian imaging,respectively.Compared with the method based on the RD image,the one based on the sparse Bayesian ISAR image is more robust,because it utilizes high order statistical information of the ISAR image.Finally,experiments based on both simulated and measured data validate that the proposed two sparse aperture ISAR autofocusing methods outperform the traditional methods in terms of robustness to noise,computational efficiency,and adaptiveness to initialization.Then,the sparse aperture ISAR cross-range scaling is studied.Two methods are proposed,which are denoted as the modified Newton methods that are based on entropy minimization and contrast maximization,respectively.First,the signal model for ISAR cross-range scaling is presented.To achieve ISAR cross-range scaling,the rotational velocity and rotational center of the target are estimated from the coefficient of the second order phase error.Note that the grid search method for 2-D optimization suffers from low computational efficiency and poor estimation accuracy.A novel modified Newton method is proposed to jointly estimate the rotational velocity and rotational center of the target.In this method,the Hessian matrix derived in each iteration is modified to be positive definite,which is achieved by reversing its negative eigvalues,so as to keep the right iterative direction,and the backtracking line search method is utilized to achieve the step,which determines the distance to go along the right iterative direction.Experimental results based on simulated and measured data of aircrafts validate that the proposed two modified Newton method based ISAR cross-range methods are robust to noise and sparse aperture,and are computationally efficient.For the sparse aperture Bi-ISAR imaging,the 2-D and 3-D signal models for Bi-ISAR imaging are built to derive the bistatic angle and analyze its effects to ISAR imaging.First,to deal with the low signal to noise(SNR)condition caused by the Bi-ISAR system,a non-coherent accumulation based method is proposed to denoise the range profiles.Note that the complex motion of the target and the time-varying bistatic angle will result in the time-varying Doppler spectrum and defocus the ISAR image.An optimal coherent processing integral(CPI)selection based on the reassigned time-frequency analysis is further proposed to find out the interval that contains nearly stationary Doppler spectrum,which is further utilized to conduct well-focused ISAR images.However,the length of the selected CPI is generally not long enough to produce the ISAR image with high cross-range resolution.Therefore,the LSM prior based sparse Bayesian method is further utilized to reconstruct well-focused ISAR images with high cross-range resolution from the selected CPI with limited length.Finally,experimental results based on the data of a cone model measured in the anechoic chamber validate the effectiveness and robustness of the proposed sparse aperture Bi-ISAR imaging method.Further,the InISAR imaging under the sparse aperture condition is studied.First,the multi-channel sparse representation is used to construct the signal model for the sparse aperture InISAR imaging,and the effect of sparse aperture on InISAR imaging is analyzed.The sparse aperture data will decrease the matched degree between different ISAR images,which will further decrease the estimate accuracy of 3-D coordinates of scatterers on the target.To improve the matched degree of ISAR images,they are jointly reconstructed from the multi-channel sparse aperture data with the usage of the multiple sparse Bayesian learning(M-SBL).Noting that the traditional M-SBL suffers from heavy computational burden,which prevents the application of the M-SBL based InISAR imaging method in practical radar systems,a sequential multiple sparse Bayesian learning(SM-SBL)is further proposed.The SM-SBL updates the unknown variables in a sequential manner to avoid the time-consuming inversion of large matrix,so as to improve the computational efficiency.After reconstructing the multi-channel ISAR images by SM-SBL,the least square(LS)method is further utilized to filter the scatterers with large estimate error out,and estimate the rotational velocity of the target simultaneously.Finally,experimental results based on simulated data validate the effectiveness of the proposed SM-SBL based sparse aperture InISAR imaging algorithm.Last chapter summarizes the thesis,and gives the next step of research.