Liquid Crystal Metasurface Enabled Hyperspectral Single-pixel Imaging

Near-infrared (NIR) hyperspectral imaging emerges as a versatile, non-destructive method for material characterisation by acquiring spatial-spectral data cubes. Yet, conventional implementations demand costly InGaAs focal-plane arrays and scanning assemblies (e.g., narrowband or slit scanners), resulting in prohibitively high cost and complexity. Here, we firstly propose a transmissive liquid crystal metasurface (LC-MS) for NIR spectral compression, which enables a computational hyperspectral single-pixel imaging (HS-SPI) system. Our streamlined design integrates a voltage-controlled LC-MS on a bucket detector within an SLM-based single-pixel imaging system, enabling spatio-spectral joint encoding and detection. The spectral image is then reconstructed by a generative deep learning model with high noise robustness. Our validation experiment presents 64×64 pixel resolution hyperspectral imaging over 261 bands (1500–1630 nm, 0.5 nm sampling), achieving a 66:1 spatio-spectral compression ratio with sub-nanometer wavelength accuracy (~0.8nm). The voltage-controlled transmissive LC-MS device facilitates seamless integration in both electronic control and optomechanical design. As an initial proof-of-concept, this approach exhibits substantial potential for further optimisation in device design and system performance, ultimately promoting cost-effective portable spectral imaging systems for transformative applications in remote sensing, field-deployable diagnostics, and mobile monitoring.