Deep learning ancient map segmentation to assess historical landscape changes

Ancient geographical maps are our window into the past for understanding spatial dynamics of last centuries. This paper proposes a novel approach to address this problem using deep learning. Convolutional neural networks (CNNs) today state-of-the-art methods in handling variety problems fields image processing. The Cassini map, created eighteenth century, is used illustrate methodology. enables us extract surfaces classes lands map: forests, heaths, arboricultural, and hydrological. evolution land use between end century andtoday was quantified by comparison with Corine Land Cover (CLC) database. For Rhone watershed, results show that arboriculture, heaths more extensive on CLC contrast hydrological network. These unprecedented new findings reveal major anthropo-climatic changes.