| Individual trees are the smallest units that make up a forest.The spatial structure and biochemical parameters of individual trees in a stand are important factors in forest resource surveys.The accuracy of forest parameters depends on the accuracy of individual tree segmentation.The traditional survey is mainly based on manual work,and LiDAR(Light Detection and Ranging)is an active remote sensing technology that can quickly obtain the three-dimensional spatial information of the target object.With the progress of technology,the acquisition cost of LiDAR data is continuously reduced,and the application and research in the field of forestry are also increasing.In particular,the detection of forest vertical structure at stand scale and individual tree scale has obvious advantages.Currently,individual tree segmentation methods can be divided into two categories according to data sources.One is CHM(Canopy Height Model)individual tree segmentation method based on rasterized canopy height model.The other is the NPC(Normalized Point Cloud)individual tree segmentation method directly based on the normalized point cloud data.However,the existing individual tree segmentation methods based on CHM and NPC have the problems of difficult individual tree extraction and low overall accuracy under high canopy density forest conditions.Therefore,this study uses UAV lidar data,taking Wanzhangshan Forest Farm in the north of Simao District in Pu ’er City,Yunnan Province as the research area,and the main tree species in study area are Pinus kesiya var.plantation.Using UAV lidar data,watershed algorithm,local maximum clustering algorithm based on point cloud,point cloud segmentation method based on stacked seed points and individual tree segmentation optimization method based on hierarchical strategy are used to segment the individual tree of Simao pine plantation with high canopy density.Meanwhile,to investigate the optimal spatial interpolation method and raster resolution size in watershed segmentation,the effect of the watershed algorithm on individual tree segmentation results under 4 CHM spatial resolutions and 3 spatial interpolation methods was analyzed.UAV high-resolution images were used as a benchmark in combination with ground survey data for visual interpretation of individual tree canopy and used as a validation,with detection rate r,accuracy rate p and F-score as evaluation indicators.The study shows that:(1)Age group differences have an effect on the accuracy of individual tree segmentation under high canopy density.The watershed algorithm,the PCS algorithm and the layer-stacked algorithm differed significantly in the segmentation accuracy of each age group in high canopy density Siemao Pine plantation,but the difference in segmentation accuracy between the PCS method based on layer-stacked and the PCS algorithm in the NPC method based on the same data source was not significant,and the segmentation accuracy of the PCS.The segmentation accuracy of the PCS algorithm was slightly higher.In single-storey forests such as young age,the segmentation accuracy of the watershed algorithm in the CHM method was better than that of the PCS algorithm and the layer-stacked algorithm method in the NPC method;in middleaged forests,there was no significant difference in the segmentation accuracy of the 3segmentation methods;while in stands with complex vertical structure such as nearmature forests,the point cloud-based local maximum clustering algorithm in the NPC single-wood method and the layer-stacked seed point-based point cloud segmentation algorithm in the NPC single-wood method were more accurate than the point cloudbased local maximum clustering algorithm.The segmentation results of the NPC single-wood method and the point cloud segmentation based on layer stacked seed points were significantly different from those of the watershed algorithm in the CHM single-wood method,and the accuracy was higher than that of the watershed algorithm.(2)Due to differences in stand structure,the results of a single individual-tree segmentation method were poor.The stratified segmentation method combining the watershed algorithm and the PCS algorithm,using age groups as a stratification index,had the highest accuracy(F = 0.73)compared to the other individual tree segmentation methods and was better than the watershed algorithm(F = 0.71),the PCS algorithm(F= 0.70)and layer-stacked algorithm(F = 0.68);the results indicate that the layered segmentation method can broaden the applicability of a single segmentation method under different stand conditions,thus improving the accuracy of individual tree segmentation.(3)However,it is often difficult to obtain information on age groups in practical applications.The advantages of the stratification method based on canopy undulation rate and height variation coefficient proposed in this paper are that it does not rely on a priori knowledge,and the age group information is replaced by canopy undulation rate and height variation coefficient in the normalized point cloud data as stratification parameters to determine the respective applicability of the 2 algorithms,combining the watershed algorithm and the PCS algorithm.The combination of the watershed algorithm and the PCS algorithm takes into account the differences in stand structure between different age groups,avoiding the problems of timeliness and accuracy of the boundaries required to obtain data on a wide range of age groups through forest management inventory data,and avoiding the large variability in stand density within the same age group when age groups are used as a stratification parameter.It also avoids the problem of large variation in stand density within the same age group when using age group as a stratification parameter,which can lead to poor stratification and thus affect overall segmentation accuracy.The accuracy of the stratification method is better than that of a single stratification method in complex stands,with some improvement in accuracy compared to the age-group based stratification method,which is based on height coefficient of variation(F=0.76)and canopy undulation rate(F=0.75).(4)The CHM generated by different interpolation methods and their different resolutions have a certain impact on the segmentation accuracy.When the CHM resolution is 0.5 m × 0.5 m,the segmentation accuracy of the watershed algorithm is the highest.Among the three spatial interpolation methods,the inverse distance weighting method(IDW)has the highest segmentation accuracy(F = 0.81),which is higher than the Kriging interpolation method(F = 0.80)and the irregular triangulation interpolation method(F = 0.77). |
