Highlights

• Chinese researchers developed mPEN, a physics-enhanced neural network that reconstructs ultrafast events at 33,000 fps with 3.6× better resolution than existing methods, using fewer photons and rare-earth-doped nanomaterials.
• The breakthrough demonstrates real-world application in food safety testing, detecting trace synthetic dyes using rare-earth upconversion nanoprobes.
• This achievement underscores China's strategic advantage in rare earth processing for advanced photonics.
• The advance in compressed high-speed imaging lowers barriers for ultrafast diagnostics in materials science and biology.
• It raises barriers for countries lacking access to high-grade rare earth intermediates.

---

Researchers led by Dr. Xing Li and Dr. Siying Wang at the Xi’an Institute of Optics and Precision Mechanics (XIOPM), working with collaborators from the National Institute for Scientific Research (INRS) in Canada and Northwestern University, report a major advance in compressed high-speed imaging, published in Ultrafast Science (IF 9.9).

The team demonstrates a new multi-prior physics-enhanced neural network (mPEN) that reconstructs ultrafast events with far higher clarity and stability than existing approaches—while enabling real-world applications using rare-earth-doped luminescent nanomaterials.

In plain terms, they found a smarter way to see extremely fast processes more clearly, using fewer photons, and with materials that China already dominates in processing.

Seeing Faster—Without Guessing Blindly

High-speed imagingfaces a fundamental challenge: reconstructing clear motion from incomplete and noisy data. Traditional deep-learning approaches often need massive training datasets and can hallucinate artifacts when conditions change. The XIOPM team tackled this by embedding multiple physical “priors”—rules drawn from physics and material behavior—directly into an untrained neural network.

Their mPEN framework integrates:

• a photoluminescence dynamics model (how light is emitted over time),
• extended sampling and sparsity constraints (to limit noise), and
• depth image priors (to correct spatial distortion).

By letting these priors cross-check one another, the system suppresses artifacts, improves spatial accuracy, and remains robust even under low-photon conditions—an Achilles’ heel for many imaging systems.

From Algorithm to Hardware

The researchers didn’t stop at theory. They built a dual-optical-path compressed imaging system using pulsed lasers, a digital micromirror device (DMD), galvanometer scanning, and synchronized CMOS cameras. One optical path encodes ultrafast motion; the other captures a reference image. Precise timing control aligns both streams, and AI-assisted reconstruction merges them into a high-fidelity video.

The result: 33,000 frames per second with spatial resolution reaching ~90.5 line pairs per millimeter—about 3.6× better than widely used COSUP-TwIST methods. Image sharpness and fidelity improved by ~1.85×, whilesignal-to-noise rose by roughly 4 dB.

Why Rare Earths Matter Here

A striking demonstration applied the system to food safety testing, detecting trace concentrations of synthetic dye using rare-earth-doped upconversion nanoprobes. These materials convert low-energy light into higher-energy emissions with exceptional stability—properties that rely on high-purity rare earth processing.

This is where supply chains enter the picture. While the paper focuses on imaging, it quietly underscores a broader reality: China’s dominance in rare earth separation and advanced materials enables downstream technologies that others struggle to scale. Sophisticated optics, sensors, and AI systems increasingly depend on rare-earth-based phosphors, lasers, and nanomaterials. Control of processing, not just mining, becomes the strategic lever.

Implications—and Cautions

Implications:

• Accelerates ultrafast diagnostics in materials science, biology, and safety testing.
• Reinforces China’s edge in rare-earth-enabled photonics and sensing.
• Raises barriers for countries lacking access to high-grade rare earth intermediates.

Limitations & Open Questions:

• Results are demonstrated on specific setups; industrial deployment will test cost and reproducibility.
• Performance gains depend on accurate physical priors—mis-specified models could limit generalization.
• The work highlights dependence on rare-earth nanomaterials, which remain vulnerable to geopolitical supply constraints.

REEx Takeaway

This study is not just an imaging milestone—it is a case study in how rare earth processing leadership translates into technological advantage. As advanced sensing and AI converge, access to refined rare earth materials increasingly determines who leads and who licenses.

Citation: Xi’an Institute of Optics andPrecision Mechanics, Chinese Academy of Sciences; Ultrafast Science, Jan. 23, 2026.

Disclaimer: This article is based on reporting from a Chinese state-affiliated research institute. Findings should be independently verified.