Research Of Parameter Extraction Of Individual Tree Based On UAV Tilt Photography Technology

As the most important tree measuring factor in forest resources investigation,single tree structure parameters play an important role in the estimation of forest growth and forest carbon pool.At present,UAV tilt photogrammetry technology has become one of the research methods to quickly and accurately obtain the parameters of single wood structure,because it has the characteristics of flexible operation,high efficiency and quickness.In this study,ginkgo(Ginkgo biloba L.)was used as the research object,and UAV tilt image data were used to reconstruct the 3D point cloud Model,which enabled the dense matching point cloud and Canopy Height Model(CHM)to be obtained in the study area.Based on the optimized CHM,this study explored the local maximum method of single wood vertices with different moving Windows(3×3 m,5×5 m,7×7 m)under different median filtering windows(3×3 m,5×5 m,7×7 m).At the same time,the single tree height was obtained according to the tree vertex position,and the first live branch height was extracted according to iterative clustering method and sparse weighted vector method,and the accuracy was verified.Then,the single tree crown was segmented by four algorithms,including Seed area growth algorithm,Watershed algorithm,Forest CAS algorithm and Point cloud-based distance discriminant clustering method,and their accuracy was compared.Finally,the stem point cloud was sliced in the study area,and the DBH of single tree was extracted by non-linear least square method,random sample consensus algorithm,average distance method and minimum circumscribed method.The extraction accuracy of single tree parameters was tested by measured values.The main results are as follows:(1)In single tree recognition based on UAV tilt image data,when 5×5 m median filtering window was combined with 5×5m local maximum method window,the effect of single tree recognition was more prominent,with an accuracy of 83.95%and an F score of88.15%;In the regression equation between predicted tree height and measured tree height,R~2was 0.99,RMSE was 2.19 m,and the fitting effect was significant.For the prediction of the first living branch of Ginkgo biloba,the iterative clustering method was more accurate than the sparse weighted vector method,and its RMSE was 0.54 m,indicating that the prediction error was within 20%.(2)In the single tree crown segmentation based on UAV tilt image data,the Seed region growth algorithm had the best extraction effect in the crown segmentation of Ginkgo broad-leaved forest.The R~2for predicting the crown area and single tree canopy was more than 0.9,and the RMSE ware 1.79 m~2and 0.99 m respectively.(3)In the extraction of single tree DBH based on UAV tilt image data,non-linear least square method,random sample consensus algorithm,average distance method and minimum external connection method underestimated the actual DBH to varying degrees.Among them,the minimum external connection method had the best result in predicting the DBH of Ginkgo broad-leaved forest,R~2and RMSE ware 0.86 cm and 9.13 cm respectively.This study proves that the UAV tilt photography technology provides a convenient,efficient and accurate scheme for the extraction of single tree structural parameters such as tree height,crown width and DBH of Ginkgo broad-leaved forest,which highlights the advantages of UAV tilt photography technology in automatic and accurate single tree recognition and segmentation,and can provide a new technical reference for intelligent and efficient forest investigation.