Multiresolution segmentation, object-oriented classification and Spot-5 imagery land use mapping.

Abstract

The object oriented classification approach represents a new paradigm to the high spatial resolution imagery processing. The use of spectral and form properties originated from the segmentation procedure allows better discrimination between objects. Fuzzy membership functions are generated from the segmented objects descriptors. The State of Parana has been currently updating its 1:50.000 land use maps by means of Spot 5 orthorectified imagery. The objective of this paper is to develop a methodology to the elaboration of land use maps by means of multi-resolution segmentation techniques and image contextual classification with the aid of fuzzy logic. In order to identify which descriptors could provide better class separability, multivariate statistic, principal components and discriminant analysis techniques were used, as a result potential descriptors were selected. Finally the classification process was achieved using those descriptors to create the fuzzy sets and the membership functions. The Land Use Map generated, including an area of 218,75 km², reached a Kappa index near to 80%, indicating the potential application of this technique nevertheless subsequent methodological adaptation might be implemented.