ABSTRACT Achieving wide‐field sensing and localized high‐quality imaging in parallel plays a vital role in various scientific research and industrial applications. However, conventional optical imaging systems still face significant challenges in balancing these two demands. Here, we propose a photon‐level single‐pixel imaging technique that can adaptively acquire the high‐quality images of dynamic regions of interest without compromising full‐field sensing capability. The technique rapidly infers the multiple attributes of dynamic objects by leveraging an enhanced image‐free sensing framework, and subsequently utilizes a pattern‐adaptive mechanism to perform regional reconstruction. Experimental results demonstrate that the proposed technique can precisely identify relevant objects while robustly eliminating the influence of interfering objects. Owing to the adaptive sampling strategy, it further achieves dynamic magnification from 1 to 12× and enables real‐time scale‐aware imaging at 16 frames per second. By allowing the on‐demand allocation of imaging resources, the technique effectively enhances the resolution of the region of interest while maintaining the full‐field sensing, offering an efficient solution for object recognition and remote monitoring in extremely low‐light scenes.
Song et al. (Wed,) studied this question.