The Study On Radar Image Reconstructing Method Based On Compressed Sensing

In the conventional radar data acquisition system, Nyquist-Shannon sampling theorem is the basic principle. Because of the restriction by this theorem, while the conventional radar improves resolution and meets the real-time requirement, it also faces the challenge of large storage capacity, fast processing speed and high sampling rate. Compressed Sensing make the sparsity of the signal as the priori knowledge, then it can reconstruct the original signal from much smaller rate than Nyquist sampling, thus effectively reduces the complexity of sensors and the sampling system. This paper, exploiting sparseness of point-like targets in the imaging space, makes Compressed Sensing theorem to radar, studies several aspects from here after.First of all, this paper elaborats the theory of Compressed Sensing; introduces the classical radar imaging theory and simulations inverse synthetic aperture radar imaging on rotator mode at the same time.Secondly this paper combines Compressed Sensing to chaotic binary-phase coded signal radar. Because chaotic phase-coded signal is pseudonoise code, then the problem of imaging is transformed to selecting dictionary and solves a convex smooth l0 norm minimization to reconstruct image. The simulation results show that, the proposed algorithm can achieve better imaging results and have the self-adaptive characteration to noise.At last, Compressed Sensing is used to the Stepped-Frequency Continuous-Wave Ground Penetrating Radar. Because it suffers slow data acquisition speed due to step-wise frequency scaning, this paper randomly selects partial numbers of frequency at every synthetic aperture then forms a non-linear decoding model and solves a convex smooth l0 norm minimization and l1 norm minimization to reconstruct subsurface image. The simulation results show that, the proposed algorithm can achieve better imaging results compared to back-projection imaging in the basis of deeply decreasing of scanning time, and smooth l0 norm minimization has better resistive-noisy ability, higher resolution than solving a convex l1 norm minimization.