In view of the problems of single data features， rough feature recognition ability， and blurred classification interval of lidar measurement technology， a ground object classification method based on compound derivative features and fuzzy Dempster-Shafer（DS）evidence synthesis theory was proposed. First， determine the recognizability of LiDAR data classification features for different types of features， and select source and derivative features with strong correlation and high discrimination in the feature space. Then， we compare the difference of the normalized difference vegetation index with green normalized difference vegetation index to ground reaction properties， and propose and construct a compound derivative feature compound normalized difference vegetation index with high identification ability. Finally， the combination of ridge-type trust allocation function performs fuzzy DS evidence synthesis and decision-making and achieves accurate classification of the ground. The experimental results show that the total classification accuracy is improved from 85.78% to 89.20%， which proves the effectiveness of the proposed method.