Year: 2026
Type: article
Abstract: This dataset provides a national vector data for built-up areas (BUA) and derived footprints envelopes extracted from the Institut National de l’Information Géographique et Forestière (IGN) SCAN 50 Historique—années 1950 raster maps. The source is a georeferenced mosaic of scanned topographic maps at 1:50,000, with a nominal ground resolution of 5 metres per pixel. The release candidate contains 5,458,764 machine-extracted BUA polygons and 128,504 derived footprint envelopes in RGF93 / Lambert-93 (EPSG:2154). Extraction combines a general binary semantic-segmentation model with a red-symbol- specific branch. Both use a ConvNeXt-Base backbone and UPerNet decoder. Model proba- bilities are thresholded and vectorised; predictions from overlapping tiles and from the two model branches are dissolved and combined. Settlement envelopes are subsequently derived from the mapped-building polygons using a 250 m contiguity rule, 100 m geometry simpli- fication, and a minimum of eight linked building polygons. The observations represent printed cartographic symbols. They are neither cadastral foot- prints nor a complete census of physically separate structures. A single polygon can represent an individual building, attached buildings, or a generalised block of built form.
Source: Apollo (University of Cambridge)
DOI: 10.17863/cam.133074
Language: English
Subfield: Ocean Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
Funder: Isaac Newton Trust
OpenAlex ID: W7204246839
FWCI:
Cites:
Cited by:
Related to: