Abstract: In view of the existing methods cannot meet the needs for accuracy and speed during
classification, this paper proposes a fast land-cover classification method for light detection and
ranging (LIDAR) data based on the non-subsampled shearlet transform (NSST) and normal
Dempster-Sharer (DS) evidence theory. At first, the NSST is used to decompose LIDAR source data
in multi-scale, and the median filter is used to reduce the noise in high frequency image from each
layer, then inverse transformation and fuse the images. Secondly, the normal probability distribution
function is built and the mass function of LIDAR data is distributed, and synthesis and decisions are
made. Experiment confirmed that the classification accuracy of the proposed method in this paper is
86.12%, while the running time is only 0.46 s. So this is a fast and high precision land-cover
classification method.

Key words: land-cover classification, light detection and ranging, non-subsampled shearlet transform;
normal DS evidence theory