关键词: 台阶线　, 边界提取　, 点云分割　, 随机森林　, Alpha Shapes　, B样条曲线

Abstract: To overcome the limitations of cumbersome operations and low accuracy in extracting bench lines from open pit mines,this study proposes a novel method based on random forest point cloud segmentation.An integrated technical frame work is developed,comprising multi-scale feature extraction,random forest-based point cloud classification,Alpha Shapes-based edge detection,and B-spline curve fitting.Specifically,multi-scale features are first extracted by computing elevation-difference descriptors (range,standard deviation,skewness,and roughness) and normal-vector difference features within neighborhood ra dii of 0.5 m,1 m,2 m and 5 m.These features are then used to train a random forest classifier,enabling precise separation of bench surfaces from slope point clouds.Subsequently,the Alpha Shapes algorithm is applied to identify edge points of the bench surfaces,which are further reconstructed into continuous and smooth bench lines using cubic B-spline curves.The pro posed approach was validated using UAV LiDAR point cloud data from the Nanihu Molybdenum Mine in Luanchuan,Henan Province,China.The results demonstrate that the method achieves a bench-line extraction accuracy (p) of 92.15%,with an average nearest-neighbor distance error (f) not exceeding 9.24 cm.These figures significantly outperform those of the profile projection method (p=81.95%, f=34.25 cm),the curvature-filtering plus Hough-transform method (p=84.33%, f=41.24 cm),and the deep-learning method PointNet++ (p=83.43%, f=27.32 cm).Overall,the proposed method demonstrates strong robustness and practical applicability. It provides a reliable technical solution for boundary extraction in complex terrains,and offers methodological support for topographic mapping in mining areas,boundary design for open-pit operations,and the development of smart mining technologies.