Study On Sparse ISAR Imaging Based On Bayesian Nonparametrics

Thanks to its unique advantages such as all-time,all-weather,long-operating range and high resolution,inverse synthetic aperture radar(ISAR)has been widely applied in military and civil fields.Under the conditions of stationary target and high signal-to-noise ratio,well-focused imaging results can be obtained by the available imaging algorithms.In practice,however,ISAR may be limited by the working mode and unable to continuously observe the target,resulting in azimuth sparse observation.Furthermore,the electromagnetic environment of the target is becoming more and more complicated,and the target echoes are contaminated by interference and noise,generating low signal-to-interference-noise ratio.Additionally,the residual phase errors after envelope alignment,together with those caused by the instability of radar system as well as propagation in atmospheric turbulence,will greatly reduce the performance of traditional imaging algorithms.Due to the sparse nature of ISAR images,high-resolution ISAR imaging problems can be converted into sparse signal reconstruction problems.Because the nonparametric Bayesian model has strong flexibility with its parameter space adaptable to data variation,it provides a new solution to high-resolution ISAR imaging in complex observation environments such as low signal-to-interference-noise ratio.Aiming at the problem of high-resolution ISAR imaging in complex environments,this dissertation establishes the sparse signal observation model and briefly introduces the ISAR imaging method based on numerical optimization and parametric Bayes.Then,it introduces the non-parametric Bayesian priors,constructs the probabilistic models,and solves the model parameters in the Bayesian framework.The related work in this dissertation provides theoretical and technical supports for the improvement of the space and air target observation capabilities of imaging radar in complex environments.The main work of this dissertation is summarized as follows:The first part describes basic principles of sparse ISAR imaging.Firstly,typical signal models of high-resolution range imaging and high-resolution azimuth imaging are derived.Then,imaging methods based on numerical optimization,such as orthogonal matching pursuit(OMP)and L1 norm optimization are discussed.After that,ISAR imaging method based on parametric sparse Bayesian learning is introduced,and results of 1-D vector reconstruction and measured data imaging are provided for performance evaluation.In the second part,two non-parametric Bayesian methods are proposed for high-resolution imaging and autofocusing of gapped data with random phase errors.First,the signal model with phase errors is constructed.Then,the Beta process prior is introduced,and the Gibbs sampling and maximum likelihood method are applied for imaging and autofocusing.In addition,the Gamma process prior is introduced in probabilistic modeling.The Maximum A Posteriori-Expectation Maximization(MAP-EM)and the maximum likelihood method are designed to obtain well-focused image.Finally,Monte Carlo simulation and imaging results of measured data verify the effectiveness of the proposed method.In the third part,a high-resolution ISAR imaging method based on gamma process prior and robust noise modeling is proposed for high-resolution imaging in complex environments with unknown noise or interference distribution.In particular,Gaussian Mixture Model(GMM)is applied for robust noise modeling,and hierarchical probabilistic model is constructed by imposing the Gamma process prior to the image vector.The model parameters are estimated effectively by MAP-EM.Finally,Monte Carlo simulation and imaging results of measured data contaminated by strong interference verify the effectiveness of the proposed method.