Fine RGB photogrametric sensor for optics/neural-network co-design

Co-design methods generally model the influence of the optical parameters only on the point spread function of the system without considering changes in light intensity that modify the signal to noise ratio of the digitized image. We include multiple noise sources in a model of imaging sensors intended for the co-design of an optical system and a neural network. We use it in co-design examples for extended depth of field in a realistic setting : with RGB images and off-axis objects.