soil erosionExtract training marks (i.e. Supervision) from OSM and integrate VGI from the humanitarian application MapSwipe, as well as some manually labeled training data to cover the whole target domain with a higher predictive accuracy. To this end, a deep learning framework called AT-CNN is proposed in which the remote sensing features detected by Deep Convolutional Neural Networks are actively transferred from a source domain to a target domain. It can fuse the knowledge of the various data sources for a more general prediction model, and is thus able to easily detect different types of buildings in urban and rural areas. 

So as to cover the entire target domain with a higher predictive accuracy. To this end, a deep learning framework called AT-CNN is proposed in which the remote sensing features detected by Deep Convolutional Neural Networks are actively transferred from a source domain to a target domain. It can fuse the knowledge of the different data sources for a more general prediction model, and is thus able to detect well different types of buildings in urban and rural areas. Read More