为充分发挥单木分割算法的应用潜力,实现单木位置的精准定位和树冠的精确划分,利用无人机载激光雷达点云数据,根据优势树种和林分密度差异选取四块人工林样地,使用单木探测率P、探测准确率R和总体精度F评价分水岭分割算法、特征点决策树算法和邻域增长算法以及点云聚类分割算法的分割精度,通过改变栅格分辨率及点云密度,进行单木分割效果的敏感性分析。结果表明:1)四种算法分割结果较好,总体F值达到089,R值总体为085,P值总体为094。林分稀疏的样地比林分密集样地分割精度更高,分水岭算法分割单木精确性和适应性最好；2)选择合适的CHM分辨率有助于提高单木分割精度,栅格分辨率为03m×03m时基于CHM分割算法分割结果均为最好；3)随着点云密度的降低,点云聚类分割算法分割精度降低。当点云密度为 100(65 pts·m-2),F值为 077,点云密度为 10(65 pts·m-2),F值为 058,F值降低019。
In order to fully explore the application potential of single wood segmentation algorithm,and to realize the precise location of single wood position,and accurate division of tree crown,the single tree segmentation algorithm of UAV LiDAR plantation is proposed.Four artificial forest samples are selected according to the difference of dominant tree species and stand density by using the point cloud data of unmanned airborne LiDAR,and the watershed segmentation algorithm is evaluated by using detection rate P,accuracy R and overall accuracy F.The sensitivity analysis of single tree segmentation is carried out by changing the grid resolution and the density of point cloud.The results show that the segmentation results of the four algorithms are good,the overall F value is 0.89,the overall R value is 0.85,and the overall P value is 0.94. It is showed that the segmentation accuracy of sparse sample plots is higher than that of dense sample plots,and the watershed algorithm has the best accuracy and adaptability；selecting the appropriate CHM resolution can improve the segmentation accuracy of single trees,and the segmentation results based on CHM algorithm are the best when the grid resolution is 0.3 m×0.3 m；with the decrease of point cloud density,the segmentation accuracy of point cloud clustering segmentation algorithm decreases.When the point cloud density is 100% (65 pts·m -2),F value is 0.77,the point cloud density is 10% (6.5 pts·m -2),F value is 0.58,F value decreases by 0.19.
YU Hai-yang, FENG Si-wei, SHEN Yang-yang, LIU Peng. Research on single tree segmentation algorithm of UAV LiDAR plantation[J]. LASER & INFRARED,2022,52(5):757~762