Distinguishing urban land-use categories in fine spatial resolution land-cover data using a graph-based, structural pattern recognition system

Abstract

This paper presents a preliminary test of a graph-based, structural pattern recognition system — known as XRAG (eXtended Relational Attribute Graph) — that might be used to infer broad categories of urban land-use from very fine spatial resolution, remotely-sensed images. XRAG allows the structural properties of, and relations between, discrete land-cover parcels to be analyzed and interpreted. Although the eventual aim is to derive land-use maps directly from remotely-sensed images, this paper employs Ordnance Survey 1:1,250 scale digital map data to provide the initial land-cover information. These data — free from the complex effects of mixed pixels, misclassification, shadowing and occlusion associated with remotely-sensed images — are used to examine the intrinsic separability of several different categories of urban land-use based on the morphological properties of, and the spatial relations between, their component land-cover parcels. In future studies, the system will be tested on real images. The current system also needs to be extended to incorporate graph-searching algorithms and graph-similarity measures, so that it can be used not only to describe the structural differences between sample areas of known land use, but also to infer land use from the spatial pattern of land cover.