Testing object oriented techniques for Corine Land Cover classification by satellite images with medium spatial resolution

Abstract

Testing object oriented techniques for Corine Land Cover classification by satellite images with medium spatial resolution.This work aims to assess the potential of segmentation and object oriented classification techniques of satellite images with medium spatial resolution, for land use/cover (Corine Land Cover, CLC) mapping. The tested procedures are assessed both in term of thematic accuracy and working time, with reference to a study area of about 4000 km 2 in central Italy. The automatic procedure is carried out by segmentation of the pan-sharpened image and by subsequent classification using membership and standard nearest neighbour functions. Results are evaluated by sample circular photoplots taken from digital IT2000 orthophotos coverage. In terms of overall accuracy, object oriented classification achieves better results than conventional on screen interpretation. The classification shows difficulties for the identification of the "mixed" classes of CLC nomenclature system; however, even in these cases the object oriented techniques provide higher producer and user accuracy than on screen interpretation. On the whole, since they are able to produce more objective and more accurate cartographic products at similar costs, the application of the tested automatic techniques seems to be preferred to the conventional on screen interpretation for satellite images with medium spatial resolution. IntroduzioneNegli ultimi decenni la necessità di disporre di banche dati aggiornate sull'uso e copertura del suolo ha assunto importanza crescente, come testimoniato dall'aggiornamento di progetti quali la classificazione di uso/copertura del suolo Corine Land Cover (Maricchiolo et al. 2004).In ambito professionale l'interpretazione visuale di immagini telerilevate resta ad oggi la tecnica operativa più utilizzata per la produzione di elaborati cartografici a supporto della pianificazione territoriale e ambientale da parte delle pubbliche amministrazioni.Recentemente l'introduzione di tecniche di classificazione object oriented (Chirici et al. 2003, Benz et al. 2004