A fast method for land-cover classification from lidar data based on hybrid dezert-smarandache model

This paper proposes a fast light detection and ranging (LIDAR) data classification method for land-cover, based on the hybrid Dezert-Smarandache (DSm) model. This proposed method improves accuracy by reducing the conflict of judging class from different data obtained by sensors. First, a generalised basic probability assignment function is built and a probability is assigned for each source image. The probability images are fused with the combination rules of hybrid DSm. Second, the integrity constraints of the hybrid DSm model are ascertained according to LIDAR data characteristics and assigned a probability for the constructed conflict sets between classes. Finally, a new rule is used to redistribute the probability of each conflict set between classes and make a final decision on classification results. The computation time of this method is decreased by at least 95% and the classification accuracy increased by 7.96% by the Dempster-Shafter (DS) method as compared with the existing method. This paper provides a high-speed and accurate method for LIDAR data land-cover classification. © 2015 ICIC International.