Ensemble Methods for Binary Classifications of Airborne LIDAR Data

Abstract

This paper presents a framework that is aimed at improving the performance of two existing ensemble methods (namely, AdaBoost and Bagging) for airborne light detection and ranging (LIDAR) classification. LIDAR is one of the fastest growing technologies to support a multitude of civil engineering applications, such as transportation, urban planning, flood control, and city 3D reconstruction. For the above applications, LIDAR data need to be classified into binary classes (i.e., terrain and nonterrain) or multiple classes (e.g., ground, vegetation, and buildings). The proposed framework is designed to enhance the generalization performance of binary classification approach by minimizing type II errors. The authors developed and tested the framework on different LIDAR data sets representing geographic sites in Germany and the United States. The results showed that the proposed ensemble framework performed better compared to the existing methods. In addition, the AdaBoost method outperformed the Bagging method on all the terrain types. However, the framework has some limitations in terms of dealing with rough terrain and discontinuous surfaces.