Research On Radar Target Imaging Using Diverse Information Based On Structured Sparse Recovery

Target imaging is the common technology in aerial defence, anti-missile equipments, space surveillance, field reconnaissance, etc. As an all weather and dim-light sensor, radar can provide high resolution images of targets, based on which the physical characteristics such as size, structure and motion can then be obtained. Driven by the new requirements of military preparation, radar imaging is also faced with new challenges and opportunities. Under this background, this thesis focuses on difficulties in research of radar imaging and the main contents and achievements are as follows.The first chapter introduces the background and significance of our research, and the difficulties and trends of radar imaging are well reviewed. The radar target imaging technologies by utilizing diversity information are summarized as well as the applications of sparse representation in radar imaging.The second chapter presents the foundational theory and algorithm. The primary principle of radar sparse imaging is firstly introduced. The sparsity of the radar echo signal is analyzed and the sparse representation model is built, based on which we introduce the conception and principle of compressed sensing radar imaging. Then four kinds of sparse recovery algorithms are reviewed and numerical simulations are perfomed to compare their performance. The results in this chapter will provide foundations for building proper structured sparse recovery algorithms later.The third chapter studies the moving target imaging in sparse random frequency hopping radar. The echo signal model of moving target is built as the parametric structured compressed sensing model, and a special iterative framework containing inside and outside iterations is applied to solve it. By utilizing this particular model and framework, the united motion compensation and imaging can then be achieved. Besides, eight criterions belonging to four kinds are proposed to evaluate the performance of the algorithm.For practical applications, the required pulse number of sparse random frequency hopping compressed sensing radar is difficult to determine in advance. In order to reduce the transmitted pulses, a dynamic strategy for HRRP generation is proposed and a complex-valued fast sequential homotopy(CV-FSH) algorithm is developed. This algorithm avoids solving a new optimization problem from scratch. The simulation results validate the algorithm is suitable for fast sequential update and requires fewer pulses than that of real-valued sequential homotopy algorithm.Considering that ISAR measurements within small rotation angle will limit the resolution of imaging, we tend to generate new ideas and methods by exploiting the prior of data correlation and sparsity. Here, two new algorithms are proposed. One is based on the combination of aperture extrapolation and block sparse Bayesian learning, which can be deemed as increasing the measurements successively. The other is to unite the prior of correlation and sparsely under the framework of sparse Bayesian learning. Simulation results demonstrate that both the algorithms can improve the imaging resolution robustly.The fourth chapter studies the polarimetric high resolution imaging of moving targets. To obtain more characteristic information of target, the polarimetric high resolution imaging consists of sparse random frequency hopping system and polarimetric measurement system. We build the parametric generalized structured compressive sensing model and solution algorithm based on time-divide and instantaneous measurement polarimetric signal model. This algorithm can jointly realize polarimetric high resolution range profile imaging and motion parameter estimation since it not only eliminates the noncoherence of polarimetric scattering matrix in the time-divide measurements, but also compensates the phase errors brought by target motion.The information extracted from small-angle Inverse Synthetic Aperture Radar(ISAR) with single-polarization antenna is not sufficient, which will lead to poor quality of image. Hence we propose polarimetric small-angle ISAR imaging algorithm. To improve the image quality, we explore and utilize the prior information such as correlations of single-polarization data as well as multi-channel data and joint sparsity. Multi-channel aperture extrapolation technique can be transformed to be the unavailable polarization measurements to exploit the correlation information. On this basis, we extrapolate the data further by using block sparse Bayesian learning method and joint sparsity. The simulated and chamber synthesized data show that the polarimetric observation can not only extract the high resolution polarimetric characteristic of target, but also can improve the resolution of image.The fifth chapter studies the radar imaging with large-angle. Firstly, we build the unified compressive observation signal model according to the linear system theory and tomographic imaging principle. The unified radial compressive observation sparse representation model is built combining the off-grid structured sparsity. We divide the compressive observation into two categories: does meet and does not meet the compressive sensing condition based on the characteristic of the equivalent compressive matrix. Moreover, we design the two different imaging algorithms accordingly.For compressive observation meeting the compressive sensing condition, we propose the strategy for radial compressive sensing data acquisition as well as hybrid imaging algorithm. This algorithm reconstructs the radial signal using compressive sensing sparse recovery algorithm firstly and then obtains two-dimensional(2-D) image with back-projection. In radial sparse recovery process, it incorporates the block sparsity of both HRRP peak position slow varying and off-grid mismatch in small angle scenario. The block orthogonal matching pursuit(BOMP) algorithm, which has low computational complexity and accurate reconstruction guarantee, is employed to improve the compressive sensing sparse recovery accuracy. This hybrid imaging algorithm has lower dimensional dictionary and better image quality than 2-D joint sparse recovery algorithm.For compressive data acquisition which does not satisfy the compressive sensing condition, e.g., narrowband observation, we propose enhanced back projection high resolution imaging algorithm. To improve the radial resolution of back projection, it exploits and explores correlation between data in different frequencies within one aspect angel, and correlation between data within small angle scenario. Also it considers the joint sparisity of HRRPs of multiple aspect angles within small angle scenario as well as correlation between HRRPs. To explore the prior correlation information involved in measurement data, the correlation is converted to the increase of observation data by bandwidth extrapolation using multi-channel technique. To explore the joint sparsity and signal correlation within one range cell, the block sparse Bayesian learning(BSBL) method is adopted exploiting the intra-block correlation. The simulation results illustrate that the proposed method could effectively improve the resolution of back projection imaging.The sixth chapter makes a summary of the research studies and main contributions in this thesis. Some open problems are also presented in this chapter.