| Popis: |
Urban growth monitoring has always been a main concern for remote sensing researchers. Spectral indices and supervised classification can be used to extract urban areas automatically by saving technicians from digitizing hundreds of polygons by hands. Multiclass classifications exhibit a very promising performance in terms of accuracy. However, such methods are expensive and labor-intensive task as they require labeling all present classes in the study area. Nevertheless, in many applications, users may only be interested in a single land class. This referred to one-class classification (OC) problem. In this paper, we compare Binary Random Forest (RF), OC-Support Vector Machine (OCSVM), and Presence and Background Learning (PBL) algorithm for the extraction of built-up areas from multi date Landsat imagery. The obtained classification accuracies show that PBL based approach provides competitive extraction results with the less effort. |
