Research On Extraction Of Single Tree Parameter Information Based On UAV Images

Whether it is ecology or forestry,the demand for forest structure information is very high.Traditional biomass estimation methods,such as logging and weighing,are often used to accurately measure biomass,but they can be expensive and time-consuming.Therefore,non-destructive methods such as remote sensing applications and lidar are increasingly used for biomass measurement.However,these methods are not as accurate as the destructive methods.Therefore,in order to find a non-invasive biomass estimation method that can properly balance the relationship between accuracy and efficiency,this paper proposes a research workflow for accurate 3D reconstruction and biomass estimation of trees based on UAV photogrammetry technology and tree quantitative structure model.The purpose of this study was to explore the feasibility of applying UAV point cloud to 3D reconstruction of a single tree.In order to test the tree modeling performance of UAV images,a total of 230 ancient and famous trees and 38 tree species in the research area were photographed in the air,followed by tree modeling.To get the best angle for the drone’s data collection,the flight was carried out during the leaf-changing season,using different tilt camera angles.After that,the single tree point cloud was filtered by noise and preprocessed by point cloud,and the tree branch structure was reconstructed geometrically.This reconstruction is based on the quantitative structure tree model,which is a geometric model that accurately simulates the branch structure and geometric shape of the tree,and can be used to extract the three-dimensional structural information of the tree and calculate the biomass of the tree.In this study,the estimated DBH of UAV point cloud and quantitative structure model were compared with measured data,and the accuracy of biomass estimation based on quantitative structure model was compared with that based on measured data.The study found that the point cloud density of trees increased with the increase of camera tilt angle.The relative root-mean-square error of DBH extraction from DBH point cloud is 9.62%,and the relative root-mean-square error of DBH extraction from two tree-fitting methods of the quantitative structure model is 6.19% and 5.86% respectively,which is not significantly different from the measured data.The biomass estimation based on volume generated by the quantitative structure model is 34.97% lower than the measured data,and the biomass estimation based on the extracted DBH and allometric equation is 27.01% lower than the reference data.Experimental results show that the dense canopy structure weakens the ability of 3d modeling of the UAV image for trees.Under such conditions,the modeling performance of quantitative structure models can be improved by means of improved flight path,repetitive flight or alternative modeling strategies.The observation accuracy of this study can meet the needs of China’s forestry survey,and it is helpful for rapid parameter extraction and biomass estimation by unmanned aerial vehicles(UAVs).It can be used for calibration and verification of biomass estimation based on satellite or measured data,and has a good prospect for popularization.