Wide-field mid-infrared hyperspectral imaging beyond video rate

Technical Summary: Wide-field mid-infrared hyperspectral imaging beyond video rate

Problem Statement
Mid-infrared (MIR) hyperspectral imaging is a critical tool for resolving chemical information in diverse samples, offering superior sensitivity compared to Raman spectroscopy due to larger infrared absorption cross-sections. However, the acquisition of three-dimensional spectral data cubes has historically been time-consuming, primarily limited by the speed of raster scanning or wavelength tuning. Existing technologies face significant bottlenecks: Fourier Transform Infrared (FTIR) systems relying on thermal sources suffer from low signal-to-noise ratios (SNR) requiring long dwell times, while mechanical scanning of interferograms limits refresh rates to minutes. Quantum Cascade Laser (QCL) systems offer faster tuning but are often restricted to single wavenumbers or reduced fields of view to maintain speed. Furthermore, direct MIR detectors (microbolometers or narrow-bandgap semiconductors) are plagued by high dark noise, low pixel counts, and limited frame rates. Consequently, achieving high-definition, video-rate (or faster) MIR hyperspectral imaging with megapixel spatial resolution and broad spectral coverage remains an unmet challenge, hindering the real-time observation of transient processes in material and life sciences.

Methodology
The authors propose and implement a high-speed, wide-field MIR hyperspectral imaging system that circumvents the limitations of direct MIR detection by utilizing broadband parametric upconversion. The core methodology involves three sequential stages:

1. Broadband Nonlinear Upconversion: A high-brightness supercontinuum (SC) laser source (1.9–3.9 µm) illuminates the sample. The transmitted MIR field is directed into a 4f imaging system where a chirped-poling lithium niobate (CPLN) nonlinear crystal is placed at the Fourier plane. A synchronized, narrow-band pump laser (1064 nm) drives sum-frequency generation (SFG). The CPLN crystal is engineered with a linearly ramping poling period (16–24 µm) to satisfy quasi-phase-matching conditions over a wide range of incident angles and signal wavelengths, enabling wide-field, broadband conversion to the visible spectrum.
2. Rapid Spectral Filtering: The resulting polychromatic upconverted field (visible range) is passed through an acousto-optic tunable filter (AOTF). Unlike mechanical filters, the AOTF is inertia-free, allowing for agile, electronic tuning of the passband with a switching time of approximately 2.8 µs. This component acts as a high-speed, programmable bandpass filter to extract monochromatic images.
3. High-Speed Detection: The filtered monochromatic images are captured by a high-definition, megapixel silicon CMOS camera capable of frame rates exceeding 10 kHz. The system utilizes an FPGA to synchronize the MIR pulse, pump pulse, AOTF tuning, and camera exposure.

A key physical characteristic of this setup is the wavelength-dependent spatial scaling factor inherent to the upconversion process at the Fourier plane. The system accounts for this by pre-determining a scaling map to numerically compensate for the radial blurring effect, ensuring accurate spatial reconstruction for each spectral channel.

Key Contributions

• System Architecture: The integration of a broadband supercontinuum source, a wide-field CPLN upconverter, and a visible-band AOTF creates a novel architecture that decouples MIR spectral imaging from the limitations of MIR detector technology.
• Speed and Resolution: The system achieves a data acquisition rate of 100 Hz for a full spectral data cube (100 bands) with a spatial resolution of megapixel scale (demonstrated at ~500 kilopixels). This represents a refreshing rate over two orders of magnitude faster than previous comparable results.
• Snapshot and Multiplexing Capability: The authors demonstrate the ability to perform snapshot operation with spatial multiplexing. By leveraging the angular dependence of phase matching, the system can simultaneously record multiple spectral channels, offering a pathway to further increase imaging speeds.
• Broadband Coverage: The system operates effectively across a broad spectral range of 2600–4085 cm⁻¹ (approx. 2.45–3.85 µm), covering key molecular fingerprint regions.

Results
The system was characterized using static and dynamic samples:

• Static Imaging: The system successfully imaged a USAF-1951 resolution target, demonstrating the ability to resolve fine features after numerical de-blurring. It was used to distinguish between two types of plastic films (polypropylene and cellulose acetate) based on their distinct absorption profiles at 2915, 3500, and 3995 cm⁻¹.
• Quantitative Analysis: The system quantified the concentration gradient of ethyl alcohol injected into a carbon tetrachloride solvent, mapping the vertical concentration distribution with high spatial fidelity.
• Dynamic Imaging (Videography): The system captured the transient diffusion process of liquid injection at a frame rate of 100 Hz. A full spectral data cube (100 bands) was acquired in 10 ms.
• Real-time Mixing Dynamics: The authors demonstrated real-time visualization of liquid mixing (ethanol and benzene in CCl₄) by selecting specific spectral bands corresponding to the unique absorption peaks of each liquid (3050 cm⁻¹ and 3350 cm⁻¹). These channels were mapped to RGB color channels to create a false-color video showing the mixing dynamics at 100 Hz.
• Performance Metrics: The system achieved a spectral resolution of approximately 15 cm⁻¹ (limited by the AOTF) and a switching time of 2.8 µs per channel. The total acquisition time for 100 spectral bands was 10 ms.

Significance and Claims
The paper claims to address the long-standing challenge of implementing video-rate (and beyond) MIR hyperspectral imaging with high pixel density. By shifting the detection to the visible spectrum via upconversion, the system leverages the superior speed and sensitivity of silicon cameras while maintaining the chemical specificity of MIR spectroscopy. The authors state that this high acquisition rate, combined with wide-field operation and broadband spectral coverage, opens new possibilities for the high-throughput characterization of transient processes in material and life sciences. The work suggests that the demonstrated speed is sufficient for real-time tracking of living organisms, fast characterization of gaseous combustion, and high-throughput sorting of biomedical samples. Additionally, the paper notes that the spatial multiplexing capability offers a route to further boost imaging rates, potentially leading to full-snapshot hyperspectral imaging. The authors also highlight that the strategy is particularly attractive for longer infrared wavelengths where sensitive imagers and fast filters are typically unavailable.