Deep Learning for Remote Sensing
N. Audebert, A. Boulch, A. Lagrange, B. Le Saux and S. Lefèvre
Published in Onera-DLR Aerospace Symposium (ODAS), 2016
This work shows how various, recent statistical techniques can benefit to remote sensing. We focus on three tasks which are recurrent in Earth-observation data analysis: multimodal classification, orthophoto rectification and aerial image segmentation. For each of them we present a novel approach based on recent developments of deep learning and discrete optimization. We assess our approaches on challenging urban, multi-sensor data-sets and establish new state-of-the-art performances. It shows that deep learning allows re-thinking the remote sensing of areas with abundant information and offers promising paths for urban monitoring and modeling.