论文部分内容阅读
针对合成孔径雷达(SAR)目标超分辨重建问题,提出了一种基于迁移学习的超分辨方法。在光学图像梯度域中联合训练超完备字典与稀疏编码映射,利用半耦合字典联系SAR图像与光学图像,寻找SAR图像在半耦合字典下的稀疏编码,并在高分辨率字典下完成重建。结合SAR图像的先验信息,使用正则化方法对SAR目标进行特征增强。所提方法在TerraSAR-X数据和MSTAR数据上进行了仿真实验,重建结果表明,相比目前的插值方法和稀疏表示方法,所提方法空间分辨率可提高0.5~1.5个像素。正则化增强结果表明,引入稀疏先验的正则化增强能够进一步提高空间分辨率并抑制杂波比,最后分析了正则化参数的选取对图像质量的影响。
Aimed at the problem of target super-resolution reconstruction in synthetic aperture radar (SAR), a super-resolution method based on migration learning is proposed. In the optical image gradient domain, the super complete dictionary and the sparse code mapping are jointly trained. The SAR image and the optical image are connected by the semi-coupled dictionary. The sparse coding of the SAR image under the semi-coupled dictionary is searched and reconstructed under the high-resolution dictionary. Combined with the a priori information of SAR images, the SAR target is enhanced by regularization method. The proposed method is simulated on TerraSAR-X data and MSTAR data. The reconstruction results show that compared with the current interpolation method and sparse representation method, the proposed method can improve the spatial resolution by 0.5 ~ 1.5 pixels. Regularization enhancement results show that the introduction of sparse priori regularization enhancement can further improve the spatial resolution and suppress the clutter ratio, and finally analyze the influence of the regularization parameter selection on the image quality.