Recently, a research team from the Xi’an Institute of Optics and Precision Mechanics (XIOPM) of the Chinese Academy of Sciences (CAS), in collaboration with the Institute National de la Recherche Scientifique (INRS) in Canada and Northwest University, developed a single-shot sCUPLI system for high-speed imaging. The study was published in the journal Ultrafast Science.
High-fidelity recovery from complex inverse problems remains a core challenge in compressed high-speed imaging. While deep learning has revolutionized reconstruction, pure end-to-end “black-box” networks often suffer from structural artifacts and high costs. To address these bottlenecks, the team proposed a multi-prior physics-enhanced neural network (mPEN).
The researchers developed the sCUPLI system by integrating mPEN with compressed optical streak ultra-high-speed photography (COSUP). This AI-powered architecture utilizes a dual-path synchronous acquisition strategy: an encoding path for temporal shearing and a prior path to record unencoded integral images. By synergistically correcting multiple complementary priors—including physical models, sparsity constraints, and deep image priors—the system effectively suppresses artifacts and corrects spatial distortion.
Technical analysis shows that the mPEN-sCUPLI achieves a spatial resolution of 90.5 lp/mm at 33,000 fps, representing an approximately 3.56-fold improvement over the TwIST-based COSUP method. Furthermore, it improves the average peak signal-to-noise ratio (PSNR) by 4 dB and enhances imaging sharpness and fidelity by 1.85 times.
To demonstrate its practical utility, the team applied the system to food safety detection. Using rare-earth-doped upconversion nanoprobes, the system successfully achieved the non-destructive and rapid detection of synthetic colorant concentrations in alcohol solutions by capturing microsecond-scale fluorescence lifetime variations.
“This research advances compressed imaging toward higher clarity and practical utility, offering promising potential for food safety and biomedical measurements,” said Professors BAI Chen and YAO Baoli. “Given its ability to achieve high fidelity and high throughput from single-shot measurements, we expect the mPEN-sCUPLI approach to be widely adopted in future quantitative detection applications.”
(published: 6 January 2026)
Figure. Schematic of the AI-powered mPEN-sCUPLI reconstruction framework. (Image by XIOPM)
Download：