Extraction Individual Tree Parameters Of Southern Plantation By Oblique Photogrammetry

Plantation forest is an important part of forest resources.It plays a key role in maintaining water and soil,conserving water resources,sequestering carbon storage,and decreasing global warming.The accurate estimation of detailed parameters such as tree height,crown width,crown area and aboveground biomass(AGB)is essential for effective management and quantitative analysis of the forest.Traditional survey methods have long periods and low efficiency,which cannot meet the needs of large-scale forest detailed surveys.The unmanned aerial vehicle(UAV)oblique photography is equipped with a multi-lens oblique camera to quickly and efficiently obtain high-resolution images from vertical and four oblique angles,which can truly reflect the three-dimensional(3D)structure information of forest and overcome the disadvantages of traditional investigation methods.Therefore,it has become a new technology of 3D real scene modelling and parameter extraction.To explore the feasibility and applicability of individual tree parameters extraction and AGB estimation based on oblique photography,this study selected Gaofeng forest farm in Nanning city of Guanxi Province as the research area.Three typical species of southern artificial forests such as Eucalyptus,Chinese fir and Acacia were taken as the research objects,a total of 25 test sites of 25 m×25 m were set up and the ground surveys were carried out.The UAV flight route was designed to collect the images of oblique photography simultaneously.The Sf M-SGM algorithm was applied to reconstruct the 3D forest scene and generated 3D dense point clouds,DSM,DOM,DTM and CHM;Based on the normalized point clouds,the AMS,MS and SRG algorithms were used to segment individual trees and extract individual tree parameters such as tree height,crown width and crown area;Aiming at the problem of low efficiency for data processing,the applicability of point cloud density was analyzed and the optimal point cloud density for obtaining the optimal parameter precision of individual tree in different stands was determined;Based on the results of individual tree segmentation,43 height variables and 10 density variables of point clouds,7 vegetation index and 5 intensity variables of DOM and extracted tree height were used to construct AGB models by using multiple linear regression and random forest and determine the optimal AGB models in different stands.The results are as follows:(1)The AMS algorithm based on normalized point clouds has achieved the best accuracy in individual tree segmentation of Eucalyptus,Chinese fir and Acacia.Compared with the MS algorithm based on fixed bandwidth and SRG algorithm,the AMS algorithm has improved the ability of individual tree detection and parameters extraction,and also decreases the phenomenon of over segmentation and under segmentation.In Eucalyptus,Acacia and Chinese fir forests,the average individual tree detection accuracy of F1-score were 0.87,which was 0.08 and 0.14 higher than the traditional MS and SRG algorithm,respectively;The phenomenon of over-segmentation and under segmentation was reduced by 6%and 5%compared with MS algorithm and reduced by 7%and 9%compared with SRG algorithm;The extraction accuracy of average tree height,crown width and the crown area was 94%,88%and 85%respectively,which were improved by 3%,2%and 12%compared with MS algorithm,and by 5%,8%and 15%respectively for SRG algorithm.(2)The results of point cloud density applicability analysis show that the point cloud density in different forest stands was proportional to the accuracy of individual tree segmentation and parameters extraction.Therefore,it is recommended to use the point cloud with the density of 100%as the input data when the fine investigation is in a small area and the requirement of parameter precision is high;In a rough survey in a large area,the requirement for parameter precision is not too high and need to pay more attention to calculation efficiency,it is recommended to use the point cloud with the density of 75%as input data.The average individual tree position detection accuracy of F1-score with 75%density point cloud in Eucalyptus,Acacia and Chinese fir forests was 0.83,and the extraction accuracy of average tree height,crown width and crown area were 93%,85%and 83%,respectively.Compared to the point cloud density was 100%,the F1-score accuracy of individual tree position detection only decreased by 0.04,and the extraction accuracy of average tree height,crown width and crown area decreased by 1%,3%and 2%respectively,which meet the requirements of a rough investigation of most individual tree parameters.(3)After individual tree segmentation,43 height variables and 10 density variables of the point cloud,7 vegetation indices and 5 intensity variables of DOM and extracted tree height were used to construct the estimation model of individual tree aboveground biomass of Eucalyptus,Chinese fir and Acacia in the survey area.The results showed that the intensity variables cloud significantly improve the accuracy of the model.For the estimation of the individual tree,AGB in coniferous forests such as Chinese fir,the random forest model had the best effect.The model accuracy R~2 was 0.94,and the root means square error(RMSE)was 41.27 kg/plant,the verification accuracy was up to 80%.For the estimation of individual tree AGB in broad-leaved forests such as Eucalyptus and Acacia,the multiple linear regression model had the best effect.The model accuracy R~2 was 0.67and 0.75,and the RMSE was 68.22 kg/plant and 61.63 kg/plant,the verification accuracy was 83%and 78%,respectively.The results show that the UAV oblique photography has great application value and development potential in parameters extraction and AGB estimation of individual trees in artificial forests.This study explores the overall technical framework from UAV route design to parameters extraction,which has important guiding significance for fine investigation and dynamic monitoring of plantation forests.