Utilizzo di Alberi Decisionali per la classificazione di aree urbanizzate

The paper describes a preliminary study on the urban classification accuracies obtained by means of the Decision Tree classifier. The study was conducted over the area of Turin (Italy), with Landsat ETM+ imagery and with an official regional map (Cartografia Tecnica Regionale) used as ground truth. In particular the variation of the accuracies was evaluated, depending on the changing of the algorithm input attributes such as the level of applied radiometric pre-processing, the considered number of classes, the temporal extent of the training set and the use of spectral indexes. Results show that overall accuracies of 80% can be achieved and that spectral indexes are the type of attribute that affect most these accuracies.