Remote Sensing and Deep Learning to Understand Noisy OpenStreetMap

The OpenStreetMap (OSM) project is an open-source, community-based, user-generated street map/data service. It the most popular within state of art for crowdsourcing. Although geometrical features and tags annotations in OSM are usually precise (particularly metropolitan areas), there instances where volunteer mapping inaccurate. Despite appeal using semantic information with remote sensing images, to train deep learning models, crowdsourced data quality inconsistent. High-resolution image segmentation a mature application many fields, such as urban planning, updated mapping, city sensing, others. Typically, supervised methods trained annotated may learn anticipate object location, but misclassification occur due noise training data. This article combines Very High Resolution (VHR) computer vision deal noisy OSM. work deals misalignment ambiguity (positional inaccuracy) concerning satellite imagery uses Convolutional Neural Network (CNN) approach detect missing buildings We propose translating method align vector strategy increases correlation between building reduce A series experiments demonstrate that our plays significant role (1) resolving issue, (2) instance-semantic (never labeled or constructed acquisitions), (3) change detection mapping. good results precision (0.96) recall viability high-resolution detection/change approach.