Using Random Forest to integrate lidar data and hyperspectral imagery for land cover classification

The elevation information derived from lidar has proven to be complementary to hyperspectral imagery which can provide the accurate description of spectral characteristics of objects. In the paper, the different ways of fusing the two distinct data sources are investigated. An integration method based on Random Forest (RF) is proposed to combine information from spectra, elevation, and their corresponding textures. The importance of each feature is scored by RF, and more useful features are chosen as inputs for RF to produce the final classification results. The experiments on hyperspectral images and Lidar data demonstrate the effectiveness of the proposed method.