3D Radar Imaging Based On Sparse Measurement

In conventional three-dimensional image formation, high range-resolution is attained by transmitting a wideband signal and high cross-range resolution is attained by increasing the observation azimuth and elevation. Consequently, large cost on collection time and data storage is required for this imaging method, which is difficult or impossible in practice. The emerging theory of compressed sensing (CS) indicates that a compressible signal can be reconstructed from just a few measurements by solving an inverse problem either through a linear programming or a greedy pursuit. Based on this, it is possible that we can recover the radar image from sparse measurements because of its sparsity property.There is an increasing interest in three-dimensional (3D) reconstruction of objects from radar measurements. This paper presents three-dimensional imaging approaches by using sparse measurements and convex optimization methods, which are connected to the CS field. Also, the quality of the recovered image is discussed. In the second chapter, we first analyze the relationship between the scene reflectivity function and the received echo of the turntable ISAR. imaging Subsequently, the elements of 3D radar imaging are presented and the principles of CS theory are introduced. Finally, the possibility of 3D radar image formation using CS theory is discussed.The third chapter of the thesis first introduces the familiar observation style which is used for sparse measurements acquisition, and then presents a practical algorithm used for 3D radar imaging. Moreover, it discusses a few different signal reconstruction methods detailedly, e.g. SSF, NESTA and SPGL1.The fourth chapter analyzes the quality of the recovered images; the impacts of various factors on the recovered image quality including sparse observation style, Signal-to-Noise ratio and reconstruction algorithm utilized are discussed. We also investigate the case where the point target is not on the grid in the k -space and show that the performance will be improved if we increase the sampling denseness of the grid in the k -space.