Semiautomatic classification procedure for updating landuse maps with high resolution optical images

This paper presents a semi-automatic procedure for the classification of high resolution images in order to obtain a fast update of land cover - land use maps. The adopted methodology can be considered as both recursive and hierarchical. It is hierarchical since it is developed as a top-down procedure starting from the main classes and moving down towards subclasses in three levels. It is recursive because in each level the classification procedure is repeated until a fixed accuracy threshold is reached by considering the results of different classifiers. Innovative aspects of the classification procedure are the use of an existing land-use map as a-priori information and an integrated land use model which indicates the probability of a certain class according to the landscape characteristics (slope, altitude and aspect). The study is carried out on six test sites located in South Tyrol, northern Italy. The procedure is applied to 6 SPOT V images (10-meter resolution. The achieved accuracy in all classes was higher than 93%.