We present in this talk the application of deep learning frameworks for satellite images. After a short introduction on deep learning, we start by presenting a couple of applications on which we are successful enough to embed the result in a production world - i.e. satellite ground segments. We then present the tools we are using (tensorflow, google cloud, spark, etc...) and point out three important aspects : the need for a new version control scheme, the need for a data visualisation tool, and the need for continuous integration. We illustrate our talk with results on automatic cloud detection within satellite images, automatic target detection, automatic generation of maps, image denoising etc... Par : Matthias Ortner Slides : https://drive.google.com/file/d/0B6vy7kY4UrXBc09IYlRldkNhRF9sSjMwaWlZeHBhNlhZZFNJ/view Talk filmé au DevFest Toulouse 2017 https://devfesttoulouse.fr