New Approaches to Classification in Remote Sensing Using Homogeneous and Hybrid Decision Trees to Map Land Cover

Decision tree classification procedures have been largely overlooked in remote sensing applications. In this paper we compare the classification performance of three types of decision trees across three different data sets. The classifiers that are considered include a univariate decision tree, multivariate decision tree, and a hybrid decision tree. Results from an n-fold cross-validation procedure show that for some datasets all the decision trees perform comparably, but for other datasets hybrid decision tree classifiers are superior because of their ability to handle complex relationships among feature attributes and class labels.