Comparison Of Airborne LiDAR Individual Tree Segmentation Methods And Analysis Of Influencing Factors

Forest resource plays an important role in ecosystem and human life.Accurately obtaining tree parameters is the main job of forest resource investigation.In recent years,the development of laser radar technology has made it possible to obtain high-precision three-dimensional structural parameters of forests.Airborne LiDAR is the main source of LiDAR data in forest inventory due to its advantages of high precision,high efficiency,and large-scale acquisition of forest structural information.At present,most of individual tree segmentation studies are conducted under specific forest conditions.There is no uniform standard for the accuracy evaluation of the segmentation method,and it is impossible to evaluate the merits of each segmentation method.In this paper,segmentation based on canopy height model,segmentation based on point cloud,segmentation based on Pit-free canopy height model,and method based on layer stacking seed point segmentation,three types of forests are coniferous forest,broad-leaved forest,and coniferous-broadleaf mixed forest.Airborne LiDAR data of different forest densities were used for individual tree segmentation of the study data.Quantitative evaluation was performed using F-Score,recall,precision,and overall accuracy,and the applicability of different segmentation methods under different forest conditions was explored.Secondly,the impact factors such as leaf area index,canopy cover,tree height coefficient of variation,tree density,slope,etc.And they were analyzed to explore their influence on each segmentation method.Provides reference for future forest inventory,selection of forest resource management options,and subsequent improvements to the segmentation algorithm.The results show:(1)There are certain differences in the results of the four segmentation methods among different forest types.In the coniferous forest plots,the accuracy based on Pit-free CHM segmentation is the best.Although the stability of the algorithm is not as good as that based on layer stacking seed points,but based on the layer stacking seed point segmentation algorithm is sensitive to the parameters,the Pit-free CHM segmentation algorithm can be preferentially selected.The precision of segmentation can reach 90%.In the broad-leaved forest sites,we found that the segmentation accuracy of the four segmentation methods was significantly lower than that of the coniferous forest sites,and the number of correctly segmented trees based on the layered seed point segmentation algorithm accounted for 77% of the total trees,and from the perspective of algorithm stability,the stability of the layer-based seed point segmentation method is also the best,and the Pit-free CHM segmentation method is easy to have an outlier due to the complexity of the forest.In the coniferous and broad mixed forest sites,the accuracy and stability of seed segmentation based on layer stacking are all good,and the stability effect based on CHM segmentation model is the worst.(2)There are also differences in the influence of the impact factors on the four segmentation methods in different forest types.In the coniferous forest plots,based on the point cloud segmentation method,with the increase of the coefficient of variation of tree height,the segmentation precision presents a downward trend.The segmentation accuracy of based on the Pit-free CHM segmentation method is best when the canopy coverage is less than 0.8.In the broad-leaf forest plots,with increasing tree densities,the segmentation accuracy of the four segmentation methods tends to decrease.When the tree density value is small,it is recommended to select the segmentation based on Pit-free CHM.On the contrary,the selection based on the layer stacking seed point segmentation method is better.Coniferous and broad-leaved forest plots,tree density,canopy coverage,and tree height variability all had significant effects on the segmentation accuracy of each segmentation method.With the increase of the coefficient of variation of the tree height,especially when tree height coefficient of variation is greater than 0.25,segmentation based on layer stacking seed points has obvious advantages.Slope has less effect on the four forest types in different forest types.