Interactive Classification of Urban Areas Using Decision Trees

Abstract

Conventionally decision trees are fixed structures for sequential classification, which are designed for certain regions and specific research questions. In heterogeneous and dynamic urban environments their broad application requires a continual change of their structure, which is time consuming and laborintensive. This study focuses on the development of a user interface to facilitate the interactive adaption of decision trees. The platform of the user interface is composed of fixed feature sets which are equally applied to all scenes. They are selected on the basis of the Transformed Divergence. The features’ thresholds are connected to controllers, which can be adapted by the user. For assessing the effectiveness of the user interface, its classification performance is compared to the one of a decision tree with fixed thresholds. By means of Landsat 7 imagery four land-cover classes are distinguished. Results show that in all analyzed test-sites the overall accuracy lies for adjusted thresholds above 80% and is by up to 33% higher than for fixed thresholds. Therefore the user interface proved to be more efficient in classifying a broad variety of scenes in urban environments than a decision tree with fixed thresholds.