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：2016
卷号：XL-3/W4
出版社：Copernicus Publications
摘要：Parcel-based classification of high-resolution images is one of the most reliable alternatives for the automatic updating of land cover/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. Qualitative information about the former land use contained in the geospatial database to be updated can also be considered, since this is often related to the current land use of the parcel. In this study, we analyse the effect that the addition of the class contained in the geospatial database as a descriptive feature has on the final classification accuracy. Since the inclusion of descriptive features as discrete ancillary data requires the employment of classifiers that are able to deal with this type of data, we chose the C5.0 algorithm, which allows us to include this type of information to create classification trees. Several accuracy degrees of the database information have been simulated in order to study the influence of this parameter on the classification accuracy. In all cases, the addition of this information as a feature increases the overall accuracy of the classification. The more precise the geospatial database information is, the more it is used in the different rules that compose the classification tree, and the more accurate the final classification is. These results have special relevance for the automatic updating of geospatial databases
关键词：Database updating; object-oriented; classification trees; high resolution imagery
