Addition of Geographic Ancillary Data for Updating Geo-spatial Databases

The geographic characteristics of the territory determine the spatial distribution of land uses and are considered as essential clues by photo-interpreters to determine the land uses. Parcel-based classification of high-resolution images is one of the most reliable alternatives for the automatic updating of land use geospatial databases. Each parcel can be characterized by means of a set of features extracted from the image, its outline, the contextual relationships with its neighbours, etc. Features derived from geographic ancillary data can be considered as descriptive information in order to characterize the objects contained in the database. Several tests have been done in order to evaluate the usefulness of different types of geographic ancillary data to improve the land use/land cover classification. The ancillary data employed are: distance maps to key geographical elements, soil maps and features extracted from digital elevation models. In this study, each database object is described with its spectral feature set extracted from the image, using a per-parcel approach, completing this information with the geographic properties. Afterwards, objects are classified using decision trees combined with boosting techniques. The assigned class is compared with the land use in the database in order to detect changes or errors in any of the compared sources. The classification results demonstrate that a significant increase in overall accuracy can be achieved by combining spectral and textural features with geographic data.