Improvement of Database Updating: Semi-automatic Urban Detection

Some years ago, the main issue of regular spatial databases updating addressed the quantity and availability of the sources. Nowadays, the abundance of satellite images moved the problem to an analytic point of view. Satellite imagery actors are currently dealing with data storage and distribution of added-value products. In this context, we present a scalable semi-automatic tool for urban detection: it is qualified with different image sources, different databases (proprietary, open source, detailed, basic etc.). The aim is not to exhaustively map buildings from a satellite image, but to give an overview of the situation regarding urban areas. It is conceived to guide stakeholders and producers throughout the updating process. The workflow presented in this article is based on existing algorithms and software resources so the application could be tested quickly on various landscapes with different sensors, in a demanding industrial context. The process is generic and adaptable, with a phase of uncorrelation, chaining a Minimum Noise Fraction transformation with a textural analysis, a learning phase, processed from an existing database, and an automatic modelling of the detected objects. The results are quantified to assess the product’s quality: 90% of the existing database is successfully recreated with less than 1% rate of potential big omissions. The method allows to detect destroyed buildings and has run in “real” updating operations, on Spot6 images (1.5 and 6 m resolution), Pleiades (1.5 and 2 m), Landsat-8 (15 m) and Sentinel-2 (10 m).