基于无人机LiDAR点云的DEM生成及其精度修正OA

Aiming at the problems of time-consuming,low efficiency and poor timeliness in traditional DEM production,we innovatively intro-duced explainable machine learning methods into DEM accuracy analysis and correction in a systematic manner,and proposed a universally ap-plicable integrated production scheme for high-precision DEM derived from unmanned aerial vehicle(UAV)LiDAR point clouds across different terrain feature regions.Firstly,we adopted the SHAP-XGBoost method to construct an error evaluation model,and analyzed the influence factors of DEM accuracy.Then,we generated the initial DEM based on the analysis results.Finally,with the help of SHAP-XGBoost again and com-bined with the key influence factors,we constructed an error correction model to achieve the accuracy correction of DEM.Experiments show that compared with point cloud density and interpolation methods,the contribution of terrain factor to the accuracy of DEM is more significant.After the point cloud density is increased to a certain extent,the effect of accuracy improvement tends to be saturated.In different terrain areas within the experimental zone,a point cloud density of 30%can meet the production requirements.The error correction amount is constructed with the variation coefficient of elevation,land-use type,slope and aspect terrain factors as the core feature,and introducing spatio-temporal cor-relation parameters simultaneously,error correction is carried out through SHAP-XGBoost,which can significantly improve DEM accuracy and effectively eliminate the phenomenon of over-limit in elevation error in individual regions with complicated terrain.