中 红 外 上 转 换 多 维 成 像 技 术 研 究 进 展(特 邀)

Significance The mid-infrared (MIR) spectral region (2 ‒ 20 μm) contains rich molecular vibrational and rotational transitions and serves as a critical window for retrieving chemical composition, molecular structure, and spatial distribution information of matter. It plays an indispensable role in biomedical diagnostics, materials characterization, and environmental monitoring. With application demands evolving from simple target visualization toward component discrimination, mechanism analysis, and multidimensional information separation, high-sensitivity acquisition of amplitude, phase, time, spectral and polarization information has become a central objective in MIR imaging. However, conventional direct MIR detection systems are fundamentally limited by high thermal noise and reduced sensitivity at room temperature, with most measurements restricted to two-dimensional intensity. Meanwhile, precise multi-degree-of-freedom control of MIR optical fields remains technically challenging, which significantly constrains efficient multidimensional sensing under weak-signal conditions. Progress Nonlinear parametric frequency upconversion has recently emerged as an effective route to overcome these limitations. By nonlinear mixing processes such as sum- and difference-frequency generation, MIR signals are coherently converted into the visible or near-infrared bands, enabling the use of mature short-wavelength detectors with low dark noise, high quantum efficiency, and fast readout, thereby supporting high-sensitivity, low-noise MIR imaging at room temperature. Moreover, programmable multi-dimensional control of the pump field allows complex modulation tasks to be shifted to the short-wavelength domain, establishing a multi-dimensional imaging framework based on pump encoding, nonlinear mapping, short-wavelength detection, and computational reconstruction. This review systematically summarizes the mechanisms, system architectures, and representative technical routes of MIR upconversion multi-dimensional imaging, with emphasis on pump-engineered multi-dimensional control. The major approaches are organized by information dimension. In the amplitude domain, pump-intensity encoding enables single-photon single-pixel imaging, nonlinear multiplexed high-speed imaging, and nonlinear pinhole imaging, improving sensitivity, frame rate, and depth of field. In the phase domain, pump-phase modulation combined with nonlinear spatial filtering supports edge-enhanced and high-order topological differential imaging, enhancing sensitivity, field of view, and spatial resolution under weak-signal conditions. In the spectral domain, broadband phase matching and rapid spectral filtering enable high-speed and high-throughput hyperspectral upconversion imaging. In the temporal domain, pulsed pump gating and temporal encoding enable single-photon ranging, three-dimensional imaging, and computational temporal ghost imaging, overcoming detector timing jitter and enabling fine transient reconstruction. In the polarization domain, time-multiplexing, inverse reconstruction, and metasurface-assisted nonlinear interference enable full-Stokes MIR polarization retrieval. These technical routes are comparatively analyzed in terms of spectral coverage, spatial and temporal resolution, field of view, sensitivity, and acquisition rate, and key performance-scaling strategies are summarized. Conclusions and Prospects Current challenges remain in the joint optimization of spectral coverage, conversion efficiency, sensitivity, field of view, and spatial resolution, as well as in long-wavelength and fingerprint-region extension, system stability, and engineering-level integration. Emerging nonlinear materials and device platforms are expanding design flexibility, while advances in phase-matching engineering and device architectures are improving performance scalability. Looking forward, developments in metasurfaces, quantum-dot structures, and nanophotonic crystals are driving MIR upconversion imaging toward highly integrated, intelligent, and chip-scale implementations. In parallel, the integration of computational imaging and deep-learning-based reconstruction is expected to substantially enhance information extraction efficiency, processing capability, and robustness in complex environments. With continued co-development of nonlinear materials, device platforms, and computational methods, MIR upconversion multi-dimensional imaging is poised to evolve into a scalable high-performance sensing framework for biochemical analysis, semiconductor nondestructive inspection, and atmospheric remote sensing, accelerating the transition of infrared imaging technologies from laboratory demonstrations to broad real-world applications.