Efficient hybrid aberration spatial-calibrated 3D network for liquid-lens-based diffractive spectral imaging

Liquid-lens-based diffractive computational spectral imaging represents one of the most promising approaches for miniaturized spectral imaging, while it suffers from image degradation caused by the coexistence of aberrations and spatial variations. To address this challenge, we propose a highly accurate and efficient hybrid aberration spatial-calibrated 3D (HASC3D) network for liquid-lens-based diffractive spectral imaging. The proposed HASC3D network integrates a hybrid aberration-spatial calibration module with a 3D convolution-enhanced U-Net architecture, significantly improving the reconstruction quality of spectral images. The custom-built imaging system, composed of a diffractive lens and a tunable liquid lens, enables nine-channel multispectral reconstruction over the 510-590 nm wavelength range with 10 nm spectral resolution. The reconstructed multispectral images achieved a peak signal-to-noise ratio (PSNR) of 24.67 dB and a structural similarity index (SSIM) of 0.67, demonstrating a clear improvement over the conventional reconstruction algorithm. With its strong correction capability and superior imaging performance, the proposed HASC3D algorithm provides a promising pathway for compact, portable, and high-performance spectral imaging technologies.