湖北林业科技

In order to evaluate different single tree segmentation methods and determine the optimal method and parameter settings for high-precision single tree segmentation using unmanned aerial vehicle LiDAR in multiple forest types,this study compared four different methods using canopy height model（CHM）and normalized point cloud（NPC）data.Watershed,variable window search, point cloud segmentation, and K-means clustering were used for single tree segmentation, and tree height parameters were extracted from nine plots of three forest types.Performance evaluation of segmentation results was conducted through actualmeasurement of plot data, and the influence of different segmentation parameters on there sults was explored. Among all the plots, the overall F-score of single tree segmentation ranges from 0.63 to 0.9, with significant differences between different single tree segmentation methods.The versatility of variable window search is good, while PCSperforms better in complex single tree segmentation. The segmentation accuracy of the two approaches is complementary in plots of different levels of complexity. Adopting differentiated segmentation strategies can help improve the efficiency and accuracy of segmentation tasks. Analyzed and summarized the segmentation performance and parameter setting schemes of various single tree segmentation methods under different forest stand conditions , providing important reference for the accurate acquisition of forest structuralparameters by unmanned aerialvehicle LiDAR technology.