Faster Multispectral Imager Based on Thin Liquid Crystal Modulator and 3D Neural Network Lattice

Abstract

Computational spectral imaging using reconstruction methods such as compressed sensing and deep learning is becoming popular. Despite the great progress, for multispectral imaging, only few expectations are realized due to various constraints. Here, a new method is proposed for multispectral sensing based on use of the following: (i) dual spectral modules, one defining the working spectral bands while the other as spectral modulator, and (ii) distributed 3D neural network algorithm. The method shows fast and accurate sensing, avoids a complicated calibration process, and can directly access any wavelength at any point. Experimental demonstration is presented using thin liquid crystal cells showing high peak signal-to-noise ratio.