Radar image can provide fine information about the target structure, which is of great importance for reconnaissance and surveillance in battlefield. Furthermore, image resolution plays an important role in automatic target recognition and intelligence gathering. With the full application of prior information, super-resolution techniques can generate high-resolution images and enhance the target structure feature from limited observations. According to the scattering characteristics of radar target in optical region, the super-resolution imaging techniques are investigated based on the sparse Bayesian learning(SBL) method.In optical region, radar returns can be considered as the coherent combination of the echoes of several scattering centers. Obviously, the sparse distribution characteristics of scattering centers meets the demand of signal sparse representation very well. Based on that, the theory and common techniques of signal sparse representation are introduced for the application of radar target imaging.According to the one-dimensional range profile imaging model, the super-resolution imaging method is studied based on sparse Bayesian learning. By analyzing the structure characteristics of the imaging model, this dissertation proposes a Toeplitz system based algorithm, which can reduce the complexity of computation and memory.Subsequently, the SBL based imaging method is generalized to the case of two-dimensional imaging. A Toeplitz-Block-Toeplitz system based algorithm is proposed with the application of the structure characteristics of the overcomplete dictionary.In the original sparse Bayesian learning algorithm, the estimation of regularization parameter is not stable when applied in radar imaging. To avoid such weak point, a decoupling algorithm is presented to select a proper regularization parameter. Simulations demonstrates that the proposed algorithm is much stable. |