| Plant phenotypic structure parameters can reflect plant structure and composition,growth and development process,physiological and biochemical characteristics.The analysis of forest phenotypic information is helpful to explore the impact of gene and environment on forest health,forestry industry and forestry sustainable development.Hippophae rhamnoides as the pioneer tree species of wind and sand fixation in Northwest China plays an extremely important role in desertification control.How to quickly and accurately extract the phenotypic information of Hippophae rhamnoides is of far-reaching significance for forest protection and management.In this study,the wild seabuckthorn forest in central and Western Wushi County of Xinjiang was selected as the research area,and two sample plots with different soil background were set up.Based on the UAV remote sensing technology and the data of height,crown,diameter and leaf area index(LAI),the height,crown area,diameter and leaf area index of single Seabuckthorn tree were extracted by local maximum,multi-scale segmentation,linear fitting and machine learning algorithm.The main results are as follows(1)Based on the canopy height model(CHM)constructed from UAV remote sensing images,the local maximum method can achieve high accuracy in extracting the tree height of sample plot 1 and sample plot2.The determination coefficient R~2of sample plot 1 is 0.9,the root mean square error(RMSE)is 0.24m,and the determination coefficient R~2of sample plot 2 is 0.76,RMSE=0.23 m.(2)Based on the digital orthophoto map(DOM),the crown area of plot 1 and plot 2 was extracted by multi-scale segmentation.The accuracy of the extracted crown area was verified with the measured value,R~2=0.65,RMSE=1.04 m~2.The extraction precision of the second crown area was R~2=0.69,RMSE=0.37 m~2.(3)The correlation crown area of Hippophae rhamnoides basal diameter was larger than that of tree height.The linear model of tree height,crown width and basal diameter was established with R~2=0.66,RMSE=1.1 cm,R~2=0.53,RMSE=1.63 cm.Among the 11 visible light vegetation indices,the correlation coefficient of determination was NRI>Ex G>Ex R=Ex GR>RGRI>NGRDI=MGRVI>CIVE>NGI>VDVI>NBI.Except for NBI,other vegetation indices were significantly correlated.(4)The input factors were 11 vegetation index and tree height,and the accuracy of Lai inversion model based on machine learning algorithm was RF>BPNN>SVM.Based on principal component analysis,the model accuracy of each machine learning algorithm is RF>SVM>BPNN.In this study,PCA processing is only effective for SVM model with RBF and linear kernel function and BP neural network algorithm with hidden layer of 1 and node of 10.It has no effect on improving the accuracy of other hidden layer nodes of BP neural network,RF model and SVM model of kernel function sigmoid.(5)Compared with the other models,the model constructed by 11 vegetation index and plant height is the best.The coefficient of determination R~2=0.82,root mean square error RMSE=0.15,the coefficient of determination R~2=0.78,root mean square error RMSE=0.21.To sum up,this study used UAV remote sensing data to extract tree height,crown area,basal diameter and leaf area index of seabuckthorn forest.The results provide an effective way to quickly obtain phenotypic information of Hippophae rhamnoides,and provide basic data and theoretical basis for selection and cultivation of Hippophae rhamnoides and forest protection management. |
