Imagen computacional híbrida de matriz espectral difractiva basada en lente líquida con una red 3D paralela fusionada...

Spectral imaging technology continues to advance toward greater miniaturization, higher efficiency, enhanced compactness, and improved tunability. In this work, we propose a hybrid liquid-lens-based diffractive spectral array computational imaging technology powered by a parallel fused spatially calibrated 3D (PFSC3D) network. This hybrid system combines a tunable liquid zoom lens with spectral array multiplexing, offering significant advantages for spectral imaging. The liquid lens enables rapid image-plane switching by electronically adjusting its focal length, while the array detector simultaneously captures focused and defocused spectral signals across multiple wavelengths. The PFSC3D network integrates a parallel spatial calibration module into a U-Net architecture enhanced with three-dimensional convolutional layers, achieving high-precision diffractive spectral array reconstruction. The reconstructed spectral range spans 480-780 nm at 20 nm spectral resolution. The reconstructed multispectral images achieve mean peak signal-to-noise ratio (MPSNR) values of 27.77, 22.68, 23.53, and 22.44 dB for the four sub-apertures of the 2×2 array, representing an average improvement of at least 1.8 dB over the conventional Unrolled Network algorithm. This approach paves the way for the development of miniaturized and embedded spectral array sensing technologies.