发表机构
School of Electronic Science and Engineering, Southeast University; School of Optoelectronic Engineering, Henan Normal University; School of Computer and Electronic Information, Nanjing Normal University; Yangzhou Jinghong Laser Technology Co., Ltd.（东南大学电子科学与工程学院; 河南师范大学光电工程学院; 南京师范大学计算机与电子信息学院; 扬州景弘激光技术有限公司）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
提出一种基于旋转光栅的合成孔径超分辨方法，通过多方向成像拓宽空间频率覆盖，结合物理先验引导的多通道深度学习网络融合图像，仅用三个方向即以λ/3.9分辨率重建物体。
AI 中文摘要
基于光栅的计算成像将高频空间信息移入可探测范围，从而实现超分辨重建。然而，固定光栅仅对特定的空间频率矢量产生有效的衍射介导频移，限制了空间频率覆盖范围。在此，我们提出一种合成孔径超分辨方法，通过旋转光栅进行重复成像，拓宽空间频率覆盖范围并扩大有效孔径。随后，一个物理先验引导的复原框架采用多通道深度学习网络，融合在不同光栅方向下获取的图像。仅利用三个光栅方向的衍射图像，所提方法以λ/3.9的分辨率重建了物体。该方法为数值孔径受限系统中光栅调制超分辨成像提供了一条实用途径。
英文摘要
Grating-based computational imaging shifts high-spatial-frequency information into the detectable range, enabling super-resolution reconstruction. However, a fixed grating produces effective diffraction-mediated shifts only for specific spatial-frequency vectors, limiting spatial-frequency coverage. Here, we propose a synthetic-aperture super-resolution method in which repeated imaging with a rotating grating broadens spatial-frequency coverage and expands the effective aperture. A physics-prior-guided restoration framework then employs a multichannel deep-learning network to fuse images acquired at different grating orientations. Using diffraction images from only three grating orientations, the proposed method reconstructed the object with a resolution of $λ/3.9$. This approach offers a practical route to grating-modulated super-resolution imaging in systems with limited numerical aperture.
