Hierarchical Land-Use Classification Using Optical Imagery and LiDAR Data

It is difficult to apply the statistical classification of optical imagery with spectral information to identify and distinguish the land-use information because the input data is highly correlated to each other event though labeled information to characterize the class distributions is typically sparse. The advent of LiDAR data with very accurate elevation information to identify and distinguish 3-dimensional informative features has given tremendous interest in the remote sensing community. In this paper, we propose new classification approach designed to integrate optical imagery and LiDAR data. The proposed method mixes the point-based classification for the LiDAR data and the statistical classification for the optical imagery. Clustering generates several class features from the elevation information of LiDAR data. The class features are used to define discriminant functions for a land class, a building class, and a tree class combined with the input data. Statistical classification generates several class features, such as the grass classes, the soil classes, the water classes, and the road classes from the spectral information of optical imagery. The class features from the LiDAR data and the optical imagery are hierarchically combined to characterize land-use information.