期刊名称：ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
印刷版ISSN：2194-9042
电子版ISSN：2194-9050
出版年度：2002
卷号：XXXIV Part 4
出版社：Copernicus Publications
摘要：Remote sensing in urban areas has been a challenger for quite some time due to their complexity and fragment with combination of man-made features and natural features. High-resolution satellite images and airborne laser altimetry data offered potential possibilities for feature extraction and spatial modelling in urban areas. Land use classification of urban areas may become possible by exploiting current high-resolution sensor data. The proposed approach incorporates spectral information from multi-spectral IKONOS images and height information from laser scanning data in hierarchical image segmentation based on semantically meaningful thresholds. By image segmentation, we obtain image objects at several levels with certain properties, which make it possible to include the spatial relations between adjacent image objects. Land cover classification and identification of image objects can be carried out mainly according to their properties. Land use classification at a higher level need to be inferred based on land cover objects and structural information at lower levels. We use Delaunay triangulation for deriving spatial relations between image objects and for structural analysis. Based on adjacency relationships of image objects, human settlements and other urban spaces are formed that create a base for land use identification as well as for structural analysis of urban areas. In this paper, the hierarchical image segmentation schema and the corresponding semantic-based thresholds are presented. To test the approach, we selected a site in a suburban area in Amsterdam, the Netherlands. The experiments show that hierarchically formed image objects are useful tools for image analysis and spatial modelling as compared to pixel-based approaches. Structural information can be derived based on hierarchical image objects, which plays an important role in land use classification in urban areas. We could cluster 704 image objects (buildings) into 82 spatial units for land use classification based on the measurement of shortest distance between adjacent buildings
