Deep Learning for Multi-Scale Mapping of Urban Land Cover from Space

Abstract

Currently, there is a high expectation in the application of machine learning methods for mapping urban land cover from space. In particular, deep learning has gained an influential role. Through investigations into the potential of deep learning, this thesis provides contributions to three aspects of urban land cover mapping on three scales: the detection of urban areas, the classification of urban land cover, and the simultaneous characterization of urban density and heterogeneity.