Résumé
This article presents a method combining marked point processes and convolutional neural networks in order to detect small objects in optical satellite images. In this setting, objects are scattered densely: the energy based formulation of a point process allows us to factor in priors to account for object interactions. Classical marked point process approaches use contrast measures to account for object location and shape, which prove limited in complex scenes. Instead, we use convolutional neural networks to learn energy terms that are more resilient to object and context visual diversity. Finally we present a procedure to learn the relative weights of prior and likelihood terms. We test our approach on remote sensing images and compare it to contrast based approaches. Code is available at https://github.com/Ayana-Inria/MPP CNN RS object detection.