Geographic object-based image analysis (GEOBIA) and data mining for urban land use classification by blocks using WorldView-2 images
ISBN 978-85-88783-11-9
Authors
1Carvalho, M.; 2Kux, H.; 3Florenzano, T.; 4Gabriella, S.
1UFF Email: marcus.carvalho@globo.com
2INPE Email: hermann@dsr.inpe.br
3INPE Email: teresa@dsr.inpe.br
4UFRJ Email: gabriella.geoufrj@gmail.com
Abstract
The objective of this study is to develop and evaluate a methodology for the analysis of WorldView-2 images based on Geographic Object-Based Image Analysis (GEOBIA) and Data Mining, to classify urban land use per block. The area under study is a section at the western part of São Paulo Metropolitan Region. Mapping land use per urban block is an important information source for managers and decision makers in urban areas. Among the land cover classes considered in this work, seven were used by the São Paulo Municipality in the official maps. Objects located within the blocks are helpful to characterize these areas. So, in order to analyze the context and the relationship among classes for the elaboration of land use mapping per block, a classification procedure was adopted – previously done with good accuracy – considering a lower hierarchical level (sub-objects) at the level of blocks (super-objects). The steps followed were: selection and sample collection at the blocks to train the classifier, choice of attributes to be analyzed by the data mining algorithm, generation and implementation of a decision tree using the DEFINIENS Developer software, for the classification of the WorldView-2 image. It is concluded that the use of the OBIA paradigm and Data Mining techniques were helpful for mapping urban land use. The Kappa index was 0.7050 and the global precision 0.7556.
Keywords
Geographic object-based image analysis; Data mining; Urban land use by blocks