Spectral reconstruction with model-based neural network for liquid crystal modulator devices

The liquid crystal modulator devices (LCMD) have become an important technique in the field of hyperspectral imaging. However, the spectral resolution and accuracy of LCMD-based imaging spectrometers are limited due to their principle. To break this limitation and promote the application of LCMD, we propose a spectral reconstruction method using model-based neural networks. The calibrated spectral transmittance of LCMD and a carefully designed loss function are used to constraint the calculation. Experiments on reconstructing both substance spectra and spectral image cubes have validated the effectiveness and superiority of the proposed method.