 A new pan-sharpening method is proposed using convolutional neural networks that adapts a simple and effective three-layer architecture for super-resolution to the pan-sharpening problem and improves performance by including several maps of non-linear radiometric indices typical of remote sensing. Experiments on three representative datasets show promising results, largely competitive with the current state of the art in terms of both full reference and no-reference metrics and also at a visual inspection. This article was authored by Giuseppe Massi, Davide Cotzalino, Luisa Verdeliva, and others.