Abstract
Upconversion detection of infrared radiation by cost-effective silicon photodetectors in visible bands has spurred a revolution in infrared imaging technology, unlocking a wide range of applications in biological imaging, optical spectroscopy, and optical data storage. Despite significant progress in upconversion detection, real-time, concurrent, complex-amplitude imaging of both phase and amplitude information, indispensable for disclosing the full signature of infrared scenes, remains a daunting challenge, impeding their widespread applications. By integrating the unique advantages of both coherent and incoherent approaches, we propose the concept of upconversion optical entropy encoding and demonstrate a video-rate infrared complex-amplitude imaging system. This is achieved by leveraging the synergistic interaction between light scattering in disordered photonic structures and lanthanide upconversion photoluminescence. By tailoring the information entropy of upconversion speckles, infrared light-field information can be captured in a single visible snapshot and explicitly reconstructed, assisted by a deep learning network, enabling infrared complex-amplitude imaging at a video rate of 25 frames per second (fps) and with high-fidelity 8-bit grayscale modulation. The high photosensitivity of the developed infrared imaging system enables a power detection limit of 0.2 nW μm(-2), three orders of magnitude lower than that of conventional parametric upconversion imaging. As a proof of concept, we demonstrate its applications in capturing video frames of natural scene images and classifying images of speed-limit signs for autonomous driving. This approach can be readily integrated with other cross-band imaging methods, paving the way for various infrared application scenarios that require video-rate, high-photosensitivity, and high-fidelity protocols.