Object-based updating of land-use maps of urban areas using satellite remote sensing

Geographical information in the form of maps is continuously subjected to change, especially in urban areas. Therefore maps have to be updated, which can be done using satellite remote sensing techniques since many satellites are in orbit today. In this paper several object-based classification and change detection techniques are investigated. An important aspect in map-updating is the translation from land cover (image domain) to land use (map domain). To study the results of several techniques, data of Landsat 5 TM (30 m), ERS 1 (30 m), Ikonos (4 m) and PHARUS (4 m) were used. The focus was on data of urban areas in the Netherlands. Classification of the images was done using spectral information, texture and in one case information on the relation between adjacent objects. Non-parametric techniques were applied because most textures appeared to be non-Gaussian distributed. The classification accuracy of Landsat 5 TM was best, second was ERS 1 and third were the results of PHARUS and Ikonos. Several reasons are given, but classifying high-resolution images is clearly more difficult than classifying low-resolution images. In case of change detection, pre-classification techniques were preferred. Although the methods can be improved, more information is required, e.g. from combining sensors or from the map to be updated. Map updating may not become fully automatic, but the job of a human operator can be made easier using the techniques investigated in this paper.