Super precision spectral encoding for laser imaging via machine learning

Abstract

Laser Imaging is an important technology nowadays, which involves using of cameras based on CCDs. However, all the colors are associated with distinct emission wavelength of laser, and CCDs not be capable of distinguishing them, rendering the RGB images captured by such cameras may not show adequate information about the laser emission wavelength. Machine Learning is a useful tool, which enables to build a model between the colors and wavelength, provided with adequate amount of data. In our work, a neural network containing residual units is used to build the model between RGB values of laser images in small blocks and the corresponding laser emission wavelength, is trained, and is used on images containing liquid crystal droplets, to make predictions of emission wavelength in every part of these images.