Research On Structured Sparse ISAR Imaging For Rotating Targets

The target with rotating parts and the spinning target are two kinds of rotating targets which can be often seen in inverse synthetic aperture radar(ISAR)imaging.The rotation of the target makes it more difficult to obtain the ISAR image,where the conventional imaging methods are invalid.The influences of target's rotation on ISAR imaging are analyzed in this dissertation.Based on the structured sparse properties of the target's high resolution range profiles and ISAR image,this dissertation proposes a series of structured sparse recovery methods to solve the problems in ISAR imaging for rotating targets.The main contents and contributions of this dissertation are summarized as follows:Firstly,this dissertation proposes the recovery methods for the joint sparse signals and the block sparse signals,which can be often seen in the signal processing field.Since the joint sparse signal has the shape of straight lines,the nonconvex hyperbolic tangent mixed norm is employed as the objective function to promote the joint sparsity.Then,the majorization-minimization(MM)method is utilized to convert the original optimization problem to a convex optimization problem,which is easier to solve.As for the block sparse signal,since the block information,i.e.,the size,the location and the shape of the block,is unknown,we partition the target scene into a number of overlapping patches.Subsequently,the local structure in each patch is learned under the sparse Bayesian framework,and the structural information is used to recover the block sparse signal.The local structure learning process and the recovering process are conducted iteratively till convergence.Simulation results show that the proposed methods can recover the structures of the signal and suppress the artificial points under low signal-to-noise ratio(SNR)with limited measurements,which indicates better performance than conventional sparse recovery methods and lays the foundation of structured sparse imaging for rotating targets.Secondly,this dissertation studies the methods to remove the micro-Doppler effect,which is caused by the rotating parts on the target.When the rotating radii is larger than half of the range resolution,the micromotion signal has the shape of sinusoids and destroys the joint sparse structures of the high resolution range profiles.In this dissertation,the micromotion signal is suppressed by promoting the joint sparsity of the high resolution range profiles.Specifically,the high resolution range profiles are assigned Bernoulli-Gaussian priors,and the entries in the same range cell share identical parameters of the priors so that the joint sparsity can be promoted.Then,the Markov chain Monte Carlo(MCMC)method is utilized to obtain the samples from the approximated posteriors,and the high resolution range profiles can be updated.When the rotating radii is smaller than half of the range resolution,the micro-Doppler effect destroys the group sparse property of the signals in the time-frequency domain,where the group sparsity can be viewed as the counterpart of joint sparsity in vector.This dissertation proposes the smoothed l2/l0(SL2L0)algorithm to eliminate the micromotion signal in the time-frequency domain by promoting the group sparsity.Then,the echoes of the main body are obtained based on its corresponding time-frequency signal.Finally,the sparse Bayesian learning algorithm is employed to accomplish the cross-range compression,so that the image of the main body can be recovered.Simulation results show that the proposed methods can effectively remove the micro-Doppler effect and achieve a clear main body image.Thirdly,this dissertation researches the adjustment method of range migration in the range compression process for spinning target.The spin of the target leads to the migration through range cells(MTRC).We incorporate the coupling phase term of the range frequency and the slow time into the observation matrix to realize the decoupling.Based on the group sparse property of the cross-range compressed signal,the alternating direction method of multipliers algorithm with mixed l2/l0 norm constraint is exploited to solve the linear equation.Since the proposed method avoids the interpolation process by using the group sparse recovery,it has better performance in the adjustment of MTRC than the methods based on the Keystone transform.Apart from the MTRC,the off-grid problem will result in the irregular range cell migration.The first-order Taylor expansion of the observation matrix with regard to the range variable is conducted,so that the modified signal model contains the grid mismatch parameters.Then,the joint sparsity-inducing Boltzmann machine prior is imposed on the high resolution range profiles to maintain the straight line structures and eliminate the irregular range cell migration.Finally,the expectation-maximization method is utilized to update the hidden variables and the model parameters iteratively.Simulation results show that the proposed methods can effectively eliminate the range migration in the high resolution range profiles.Fourthly,this dissertation studies the problems about the autofocusing and the cross-range scaling in ISAR imaging of the spinning target.A novel signal model is proposed to describe the initial phase errors and the high-order phase errors simultaneously.The block sparsity constraints are introduced in the imaging process.In other words,we take advantage of the correlation of adjacent scatterters to maintain the weak scatterers,so that a more accurate image can be achieved.Based on the recovered image,the initial phase errors and the high-order phase errors are estimated by maximizing the likelihood.Since the angular velocity of the target can be obtained through the high-order phase errors,we can acquire the cross-range resolution,which is vital in cross-range scaling of the target's image.Simulation results show that the proposed method can achieve autofocus and obtain a clear image.Besides,the imaging result can display the true size of the target through cross-range scaling.