A comparison on multiple level features for fusion of hyperspectral and LiDAR data

Remote sensed images contain a wealth of information. Next to diverse sensor technologies that allow us to measure different aspects of objects on the Earth (spectral characteristics in hyperspectral (HS) images, height in Light Detection And Ranging (LiDAR) data), we also have advanced image processing algorithms that have been developed to mine relevant information from multisensor remote sensing data for Earth observation. However, automatic interpretation of remote sensed images is still very difficult. In this paper, we compare multiple level features for fusion of HS and LiDAR data for urban area classification. Experimental results on fusion of HS and LiDAR data from the 2013 IEEE GRSS Data Fusion Contest demonstrate that middle-level morphological attribute features outperform high-level deep learning features. Compared to the methods using raw data fusion and deep learning fusion, with the graph-based fusion method [4], overall classification accuracies were improved by 8%.