Classification of VHR optical data for land use change analysis by scale object seletion (SOS) algorithm

This work presents the main outcomes of a land use change detection analysis by means of a classification algorithm based on very high resolution (VHR) optical images. The satellite data we used were captured by the sensor on board of IKONOS and GeoEye-1, on January 2002 and June 2012, respectively. Land use map at each day has been obtained merging the results of a supervised multispectral per-pixel classification and an unsupervised hierarchical segmentation aiming at classifying the objects in the VHR images selecting the spatial scales which maximize the final classification accuracy. The change detection land use analysis has been performed in post classification phase, comparing the resulting land use maps. The implemented classification architecture, called Scale Object Selection (SOS), allowed to obtain an overall classification accuracy higher than 97%, and a K-coefficient of about 0.95.