Zusammenfassung
Summary
Land cover has undergone tremendous changes in the last century, with strong consequences on the environmental balance, therefore the classification of land cover using Earth Observation data is very important. Recently, the widely used pixel-based approach for land cover classification was questioned through the emergence of the object-based approach. The main objective of this paper is to evaluate the performances of pixel-based and object-based approaches through three classification methods, Random Trees, Decision Tree and Kappa Nearest Neighbor, for land cover classification using Landsat satellite imagery. Our results suggest that using methods based on decision trees (Random Trees or Decision Tree) the accuracy values are superior in the object-based approach compared to pixel-based approach. Instead, using KNN classification, accuracy values are superior in the pixel-based approach.