Study On Cunninghamia Lanceolata Volume Estimation Based On UAV Image

At present,the calculation of forest volume is still obtained through field investigations,which consumes a lot of manpower and material resources.Cunninghamia lanceolate is an important fast-growing timber species in South China.How to rapidly calculate its forest volume is an important task of forestry.Remote sensing technology has been gradually used in the estimation of forest volume due to its characteristics of large amount of information,wide coverage and short acquisition period.Most of researches are with respect to optical remote sensing images taken by satellites to construct the volume inversion model,but the accuracy is relatively low.Recently,with the rapid development of UAV technology,it has gradually be-come a new research trend to estimate forest stock using UAV remote sensing images.In this paper,the Huangfengqiao State-owned Forest Farm in Hunan Province was utilized as the research area.The 30 plots of the Cunninghamia lanceolate in the study area were used as research objects,and the data of the DBH,tree height and crown width of the field were measured.Retrieved image data for each sample plot was obtained from multiple angles using UAV.The image data of each sample area was spliced,registered,and corrected to obtain orthophoto images.The three-dimensional point cloud data was produced by the application of aerial trian gulation,a beam method LAN adjustment,and null three encryption operations.On the basis of the orthophotos and 3D point cloud data,the average height information of crown,tree,and stand was extracted.This helped construct a multiple linear regression model,logarithmic model,and least squares regression model.By comparing and analyzing the three models obtained,a method for inversion of Cunninghamia lanceolate volume based on UAV remote sensing image was obtained.The specific research contents and results were as follows:(1)UAV acquires Orthophoto Images and high-density point cloud data.The route is set by Altizure software to collect image data of the unmanned aircraft field,and image data with quality assurance is obtained.The approximate nearest neighbor algorithm in the Pix4D software is utilized to realize the same-name point matching recognition of the single image of the drone,and high-quality orthophotos and high-density point cloud data can be generated.(2)The tree-crown can be accurately extracted.The canopy crown is extracted by the objectoriented method.When the segmentation parameters are selected,the local change rate LV is calculated by the ESP scale calculation tool,and the seg-mentation parameters are quantized,which avoids the process of determining the optimal segmentation parameters by multiple experiments.Then,spectral information,texture information,and vegetation index factor are used to extract crown width.The obtained crown width and number of trees are better,and the two-sided t-test shows no significant difference.(3)The accuracy of extracting the average height of trees is high.The average height of the stand of the plot is extracted by extracting the height of some trees,and taking the arithmetic mean as the average height of the whole plot.Firstly,the difference between the ground point cloud and the tree point cloud is judged.The single tree with the ground point cloud data nearby is selected by Terrasolid software,and the maximum and minimum values of the elevation coordinates of the point cloud data in the area are selected,and the difference is taken as the The height of the trees,the average value of multiple trees as the average height of the stand.For the plots with high canopy closure and low ground point cloud data,we can combine with the measured ground information points,to obtain the average height of the stands for the high canopy closure plots.The average accuracy of the average height of the forests extracted from 30 plots was 70.14%,the highest precision was 92.12%,and the overall accuracy was 83.98%.(4)The inversion model of Cunninghamia lanceolate volume was constructed.In this paper,volume is taken as dependent variable.The tree-crown,tree number and average stand height are taken as independent variables.Three different models are constructed,namely,multiple linear regression model,logarithmic model and least square regression model.The fitting effect of least squares regression model is the best among the three models,and R2 is 0.73.Followed by multiple linear regression model,and R2 is 0.57.The worst is logarithmic model,and R2 is 0.51.The least squares regression model fitting determination coefficient is 0.73.The predicted root mean square error is 11.78 D:m3/hm2,and the relative error is 20.07%.The multi-ple linear regression model fitting determination coefficient R2 is 0.55.The root mean square error is 10.89 m3/hm2,and the relative error is 28.47%.The logarithmic model fitting determination coefficient R2 is 0.47.The root mean square error is 18.75 m3/hm2,and the relative error is 37.69%.Combined with the research results,it is feasible to use UAV remote sensing technology to retrieve the volume of Cunninghamia lanceolate.Especially for small scale plantations with low canopy density,the data of forest resources can be rapidly obtained by UAV,which plays an important role in forestry monitoring and survey data acquisition.