


Deep learning based polygonal object outline extraction from airborne imagery
Weiqin Jiao is a PhD student in the department Scientific. (Co)Promotors are prof.dr.ing. C. Persello; prof.dr.ir. M.G. Vosselman & dr.-ing. H. Cheng from the faculty of Geo-Information Science and Earth Observation (ITC), University of Twente.
High-quality topographic maps are essential for urban planning, infrastructure management, environmental monitoring, and many other applications. However, producing and updating these maps still requires substantial manual work. This thesis investigates how deep learning can support more automated topographic mapping from high-resolution aerial imagery. The research focuses on generating geographic objects directly as vector polygons, rather than only as raster images. It progresses from building outline extraction, to road polygons with complex geometry and connectivity, and finally to complete multi-class vector basemaps. Across these tasks, the goal is not only to achieve accurate object recognition, but also to produce geometrically regular, compact, and topologically consistent map features. The thesis develops several deep-learning methods and task formulations that improve the efficiency and quality of polygon generation. The final work further addresses seamless map generation, where neighbouring polygons share boundaries and gaps or overlaps are avoided. Overall, this research contributes towards more automated, reliable, and scalable workflows for creating and updating vector topographic maps.