中文摘要：

基于正则化变分的框架，提出了SAR图像目标超分辨的变范数算法。考虑目标在成像区域中的稀疏特性，利用广义高斯分布对目标区域的幅度进行建模，在Bayes估计的框架下，推导了l_p范数约束的正则化变分模型和广义高斯分布形状参数的关系。采用迭代的方法在逐次估计真实图像的过程中，将P的取值与逐次估计结果相关联，逐步估计目标区域分布的形状参数，并修正l_p范数的具体形式，由此得到变范数的正则化模型。该方法克服了通过经验选取P值的局限，以及由观测数据估计P值的误差。仿真和实测SAR图像的处理结果表明了该方法的有效性。

英文摘要：

Based on the framework oof regularization variation, we proposed a superresolufion algorithm of SAR image targets using variable norms. Considering the sparseness of scatterings in the scene, the generalized Gaussian distribution（GGD） is used to model the amplitude of the targets data. Following Bayesian framework, we deduced the relationship between l_p norm regularization and the shape parameter of GGD. Under an iterative scheme, we associated the parameter p with the estimated results of each iteration step. The estimated parameter p was then used to modify the model and to construct a superresolution model with variable l_p norms. Our method overcomes the shortcomings of the fixed le norm regularization model, which selects the parameter p by experience. Finally, we demonstrated the performance of our method on simulated and real SAR scene. Experiments show that the method selects parameter p automatically in an iterative way, and produces super-resolved SAR images.