Object-oriented classification of QuickBird images for forest mapping

Abstract

Manual photointerpretation in GIS environment is the approach commonly used to derive forest thematic maps by satellite images. Forest maps produced by supervised or unsupervised pixel-based methods are usually characterized by systems of nomenclature not enough detailed to be used as usefull tools supporting forest planning. The use of more complex legends make classification errors higher, usually over standard acceptability. Such problems still limit the operative diffusion of thematic maps produced by supervised and unsupervised pixel-based classifications of remotely sensed images. The last generation of very high resolution satellites and the development of object-oriented classification techniques seem to be able to bypass such problems. For such a task in a study area of 1370 ha (Sasso Marconi, Bologna, Italy) the efficacy of an object-oriented classification method based on multiresolution segmentation of QuickBird satellite images is tested. Achieved map is then contrasted with pixel-oriented map produced with a standard supervised classification of the same image.