Multispectral Camera Calibration Using Convolutional Neural Networks

Abstract

A methodology for multispectral camera calibration using convolutional neural networks is presented. RGB images were captured from the multispectral camera for each of the standards, the samples are taken under the same lighting conditions and with the same capture angle. The images are fragmented into small matrix sizes added to a specific class, and saved with a special label to distinguish it from the entire class database, the same process takes the remaining 7 Lucideon Std color tiles. One of the tiles will correspond to a particular class with an equal dimension for all classes. Finally, based on the presented methodology, it is possible to calibrate the camera with respect to the references.