🔬 optics

High-speed hyperspectral 3D ghost imaging LiDAR

This paper presents a high-speed hyperspectral 3D ghost imaging LiDAR system that utilizes stochastic broadband lasers and spatiotemporal encoding to simultaneously achieve rapid 3D structural mapping and high-resolution spectral chemical identification, overcoming the speed limitations of conventional wavelength-scanning methods.

Original authors: Jing Hu, Wending Huang, Tianjian Lv, Ming Yan, Zhaoyang Wen, Zijian Wang, Yuan Chen, Zhuoren Wan, Mei Yang, Heping Zeng

Published 2026-04-24

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Jing Hu, Wending Huang, Tianjian Lv, Ming Yan, Zhaoyang Wen, Zijian Wang, Yuan Chen, Zhuoren Wan, Mei Yang, Heping Zeng

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you have a super-powered flashlight that doesn't just see where things are, but also instantly knows what they are made of, all while moving at the speed of light. That is essentially what this paper describes: a new kind of "smart radar" called Hyperspectral 3D Ghost Imaging LiDAR.

Here is the breakdown of how it works, using some everyday analogies.

1. The Problem: The Slow "Flashlight"

Traditional LiDAR (used in self-driving cars) is like a laser pointer. It shoots a beam, waits for the bounce, and calculates distance. It's great at making 3D maps, but it's "blind" to chemistry. It can tell you a wall is there, but it can't tell you if the wall is made of wood, plastic, or toxic gas.

To see the chemical makeup (spectral info), old systems had to use a "slow scanner." Imagine trying to identify a painting by looking at it through a prism, but you can only look at one tiny color at a time. You have to scan the whole painting, then change the prism, scan again, and repeat. This is too slow for moving cars or fast-moving robots.

2. The Solution: The "Stochastic Strobe"

The researchers built a system that acts like a super-fast, chaotic strobe light.

The Light Source: Instead of a steady laser, they use a "supercontinuum" laser. Think of this as a light bulb that flickers wildly and randomly, changing its color pattern billions of times per second. Every single flash of light is unique, like a fingerprint.
The "Ghost" Trick: This is the coolest part. They don't use a camera with a million pixels (like your phone). Instead, they use one single light sensor (a "bucket detector").
- The Analogy: Imagine you are in a dark room with a friend. You throw a ball at a wall covered in different colored stickers. You can't see the wall. But, you have a friend holding a "magic mirror" (the reference arm) that records exactly how the light looked before it hit the wall.
- By comparing the "messy" light that came back from the wall with the "magic mirror" record, a computer can mathematically reconstruct exactly what the wall looked like, even though you only used one eye (one sensor). This is "Ghost Imaging."

3. How It Sees in 3D (The Echo Location)

While doing the chemical analysis, the system also acts like a bat.

The light is split into a "train" of tiny sub-pulses, like a row of runners spaced out by a few inches.
When these runners hit an object, they bounce back.
The system measures exactly how long it takes for each runner to return.
Result: It builds a 3D map of the object's shape (depth) at the exact same time it analyzes the color.

4. The Speed: The "Bullet Train"

The most impressive part of this paper is the speed.

Old Way: Like a snail crawling, scanning one color, then another.
New Way: Like a bullet train.
- The system fires 60.5 million pulses every second.
- It captures 1.8 billion data points per second.
- It can measure distance with the precision of 0.02 millimeters (thinner than a human hair) in just 10 microseconds (faster than a blink).

5. Why Does This Matter? (The Real-World Magic)

Because this system is so fast and sees both shape and chemistry, it can do things that were previously impossible:

The "Chemical X-Ray": If you point this at a forest, it won't just see "trees." It can instantly tell you which trees are healthy, which are dying, and which are different species, all while the car is driving by.
Hazardous Material Detection: It could fly over a factory and instantly spot a leak of toxic gas (like benzene or ethanol) by its unique "chemical fingerprint," even if the gas is invisible to the naked eye.
Industrial Inspection: It could scan a car on an assembly line, checking if the paint is the right chemical mix and if the bumper is the right shape, all in a split second.

Summary

Think of this technology as upgrading from a black-and-white sketch (standard LiDAR) to a live, high-definition, chemical-sensing movie that plays at the speed of light. It solves the old problem of "speed vs. detail" by using a chaotic light source and a clever math trick (Ghost Imaging) to see everything at once.

1. Problem Statement

Light Detection and Ranging (LiDAR) is critical for autonomous systems and industrial metrology. However, existing systems face a fundamental trade-off:

Conventional Hyperspectral LiDAR: Relies on mechanical wavelength scanning or spectrometer-based detection. This sequential acquisition limits imaging speed, making it unsuitable for dynamic scenes.
Computational Imaging: While fast, traditional methods often lack depth resolution or rely on Digital Micromirror Devices (DMDs) that bottleneck speed.
Gap: There is a lack of systems capable of simultaneously acquiring high-speed, co-registered 3D spatial structure and broadband spectral information (chemical identity) in real-time.

2. Methodology

The authors propose a unified Hyperspectral 3D Ghost Imaging LiDAR system that integrates Time-of-Flight (ToF) ranging with Spectral Ghost Imaging (SGI) using a single-pixel detector.

Core Principle: The system uses a stochastic broadband laser source (supercontinuum) where each pulse possesses a unique, random spectral profile. This acts as a random encoding basis for spectral reconstruction.
Dual-Arm Architecture:
- Detection Arm (Spatial & Depth): The laser pulses pass through a Spatio-Temporal Encoder (STE). The STE (using a mirror and glass plate) splits a single pulse into an array of sub-pulses with fixed temporal spacing ( $\Delta T$ $Δ T$ ) and spatial separation ( $\Delta x$ $Δ x$ ). These sub-pulses sequentially illuminate the target.
  - Depth ( $z$ ): Determined by the Time-of-Flight ( $t_i$ ) of the echo.
  - Lateral Position ( $x, y$ ): Determined by the sub-pulse index and scanning mechanism (galvanometric scanner or optical multiplexer).
- Reference Arm (Spectral): A small fraction of the light is sent through a Dispersion-Managed Fiber (DMF). This performs a Time-Stretch Dispersive Fourier Transform (TS-DFT), mapping the pulse's spectrum into a temporal waveform.
Reconstruction Algorithm:
- 3D Geometry: Reconstructed via ToF analysis of the echo train.
- Spectral Data: Reconstructed via Ghost Imaging. The system correlates the integrated intensity of the reflected pulses (detection arm) with the time-stretched spectral waveforms (reference arm) using a second-order correlation function. This allows for pulse-resolved spectral recovery without a spectrometer.

3. Key Contributions

Unified 4D Sensing: First demonstration of a system that simultaneously recovers 3D spatial coordinates ( $x, y, z$ ) and spectral information ( $\lambda$ ) in a single time-domain framework.
Elimination of Mechanical Scanning: By using spatiotemporal encoding and stochastic spectral patterns, the system removes the need for mechanical wavelength scanning or large detector arrays.
Single-Pixel Detection: Achieves high-dimensional data acquisition using a single high-speed photodetector, significantly reducing system complexity and cost.
Pulse-Resolved Acquisition: Enables spectral reconstruction on a per-pulse basis, allowing for real-time processing at the laser repetition rate.

4. Key Results

Imaging Speed:
- Achieved a line-scanning rate of 60.5 MHz.
- Point acquisition rate (PAR) of 1.8 GHz (for 1D scanning) and up to 3.3 GHz (for 2D scanning).
Precision:
- Ranging Precision: 0.02 mm (Allan deviation) within a 10 $\mu$ s integration time.
- Spectral Resolution: 1.4 nm over the 1100–1250 nm band.
Performance Validation:
- 3D Imaging: Successfully reconstructed 3D point clouds of "ECNU" letters with depth deviations fitting Gaussian distributions (FWHM ~0.25 mm).
- Spectral Fidelity: Retrieved spectra matched commercial FTIR spectrometer data with high correlation.
- Chemical Identification: Demonstrated the ability to distinguish between four chemical samples (benzene, ethanol, toluene, PMMA) placed in front of the targets.
- Data Analysis: Principal Component Analysis (PCA) on the hyperspectral data cube achieved 98.7% variance retention, clearly separating chemical clusters, whereas single-wavelength analysis failed to distinguish them.

5. Significance and Applications

This work represents a paradigm shift in remote sensing by overcoming the speed bottlenecks of traditional hyperspectral LiDAR.

Dynamic Monitoring: The high acquisition speed enables the imaging of moving objects and dynamic processes, which was previously impossible with spectral LiDAR.
Material Identification: The ability to perform simultaneous 3D mapping and chemical identification is crucial for:
- Precision Agriculture: Assessing crop health, species, and morphology simultaneously.
- Environmental Monitoring: Detecting and mapping pollutants in real-time.
- Industrial Inspection: Closed-loop control for manufacturing based on both geometry and material composition.
- Hazardous Material Detection: Identifying specific chemicals in complex 3D environments.
Future Scalability: The framework is extensible to single-photon detection (for low-light/biomedical applications) and mid-infrared frequencies (for molecular fingerprinting), offering a flexible platform for next-generation sensing.