Research On Single Tree Segmentation And Tree Height Estimation Method Based On UAV LiDAR

Forest structure information,such as tree height,crown diameter,diameter at breast height(DBH)and canopy volume,are not only important indicators for assessing forest growth status,spatial structure and ecological function,but also the basis for analyzing the carbon balance of global forest ecosystems.Among them,the height is an important parameter of forest resource survey,which is often used for site quality and tree growth status evaluation,tree volume and biomass estimation.Light Detection and Ranging(Airborne LiDAR)can actively emit laser energy pulses,penetrate the dense vegetation canopy to a certain extent,and quickly obtain high-precision three-dimensional spatial information of the canopy and forest terrain,so as to provide support for accurate estimation of forest structure parameters.The Canopy Height Model(CHM)is one of the most common methods to estimate the height of single-tree.However,CHM generated by lidar in forest with large slope is often distorted,which will reduce the accuracy of single tree.In response to this problem,this study took coniferous forests in areas with large undulations near Fujiang Village in Xing'an County,Guangxi Province,and used Unmanned Aerial Vehicle LiDAR(UAV LiDAR)point cloud data in the study area.Proposed a new tree height estimation method to improve estimation accuracy,which combines CHM and Digital Surface Model(DSM).The main research contents and conclusions were as follows:(1)Preprocessing of laser point cloud in forest areas.Firstly,the spatial index of point cloud data was established by using octree,and the denoising method based on distance statistics was used to remove the noise of point cloud data;Secondly,the forest ground point filtering was tested and compared with slope filtering,multi-level moving surface fitting filtering and progressive encryption triangulation filtering,and the optimal method was selected to filter the study area;Finally,the inverse distance-weighted interpolation method was used to generate DEM and DSM,and the two were subtracted to obtain CHM,then the pits on the CHM was estimated by using the statistical pit fill method.(2)Segmentation of single tree based on marker control watershed.Firstly,based on the CHM after pit removal,the local maximum algorithm was used to detect the tops of the tree;Secondly,combined with the tree tops,the Mark-Controlled Watershed Segmentation(MCWS)method and the Normalized Cut(Ncut)method were used to segment the single tree respectively;Finally,the segmentation accuracy of MCWS and Ncut was compared.The results showed that compared with Ncut,the number of missing segmentation and under-segmentation of canopy was less and the overall accuracy was higher by using MCWS,which is more suitable for areas with large terrain slope.(3)Estimation of single tree height combined with DSM.In order to solve the problem that error estimation of the signal tree height caused by CHM distortion in area with large slope,this paper used the method of combining CHM and DSM to estimate the height of signal tree.Firstly,extracted the crown contour polygon obtained by the marker control watershed algorithm to DSM,and marked it as the marking area;then compared each pixel in the marked area with its neighborhood,and selected the the largest pixel value point in this area as the tree top,the tree height was obtained by using the height of the tree top to subtracte the ground elevation interpolated by Delaunay triangulation;Finally,in order to verify the tree height estimation method proposed in this paper,the experimental areas with slope of 32°,25°and 15°were tested respectively.The results showed that R~2 of estimated tree height in CHM and the measured data were0.84,0.85,and 0.87,respectively.RMSE was 1.48 m,1.41 m,and 1.58 m respectively.While,the accuracy of combining DSM and CHM was improved,R~2 is 0.92,0.91,and0.93 respectively,and RMSE was 0.93 m,1.02 m,and 1.16 m respectively.The results showed that the method in this paper could effectively improve the estimation accuracy of single tree height in areas with large slope,and it was of great significance to improving the accuracy of forest resource investigation.