| As an important economic forest tree species,Camellia oleifera has high medicinal and edible value.In recent years,the demand for Camellia oleifera in China has been increasing.Therefore,it is of great significance to scientifically manage Camellia oleifera and accurately estimate its yield.The traditional camellia production survey is mainly about picking fresh fruits and weighing them.With the development of remote sensing technology,drones are widely used in the field of forestry.The high-resolution images obtained by drones can clearly show the color and texture of ground objects.The information contained in the images was extracted and the response variables were statistically analyzed,which provided an effective method for the inversion of tree parameters and the estimation of Camellia oleifera yield in a small-scale range.This paper takes the Camellia oleifera forest in Mingyue Village,Changsha County as the research area,selects 85 Camellia oleifera trees to measure the crown width,tree height and yield,and uses Unmanned Aerial Vehicle(UAV)photogrammetry and 3-dimensional laser scanning technology to obtain digital orthophoto images,densely matched point clouds and Laser point cloud data.First of all,in view of the limitation of scale on the spatial structure of ground objects and the problems that traditional watershed segmentation is prone to over-segmentation of the canopy,a canopy segmentation method based on multi-scale markers to optimize the watershed is proposed to achieve accurate identification of Camellia oleifera trees and rapid crown width.extract.Secondly,remote sensing variables,tree height and canopy volume are extracted based on Digital Orthophoto Map(DOM)and densely matched point cloud,the tree height is extracted by local maximum method,and the tree height is extracted by kriging method,inverse distance weight method,natural adjacent point method and filtering triangulation method respectively.Camellia oleifera crown volume.Finally,different feature sets were constructed by using the extracted variables,and the yield of Camellia oleifera was estimated by multiple linear regression,random forest,and k-nearest neighbor model,and the feasibility of extracting Camellia oleifera parameters and estimating Camellia oleifera yield by using UAV oblique photography was discussed.The research analysis can draw the following conclusions:(1)The canopy segmentation method of optimized watershed with multi-scale markers is better than the single watershed segmentation result.The parameter settings of multi-scale marker watershed segmentation include tree crown marker iterative growth scale and watershed segmentation threshold scale.The optimal iterative growth scale determined by the Johnson index is 20,the watershed segmentation threshold scale is 85,and the extracted value of Camellia oleifera crown width under the optimal scale The R2 of the measured value is 0.69,the relative error between the extracted value of the canopy area and the visual interpretation reference value is 13.6%,and the overall accuracy of single tree identification is 87.1%,which is 30.4%higher than the traditional watershed segmentation method.To a certain extent,it can solve the problem of over-segmentation that is easy to occur when traditional watershed segmentation is applied to high-resolution images.(2)The filter triangulation method is the best method to obtain Camellia oleifera canopy volume,and its average relative error(31.54%)is better than the inverse distance weighting method(36.73%),the kriging method(37.04%),and the natural neighboring point method(38.54%).The relative errors of the volume obtained by the four interpolation methods are all between 30 and 40%,and the volume calculation results are generally larger than the measured values.(3)Remote sensing variables and single tree parameters are used together as characteristic variables to estimate yield,which can improve the accuracy of yield estimation.The estimation results of its multiple regression,random forest,and k-nearest neighbor models are better than the estimation results of using remote sensing variables or single tree parameters alone.The R2 of the three models were 0.35,0.64,and 0.25,and the RRMSEs were 30.97%,28.69%,and 31.36%,respectively.(4)Among the several methods for estimating the yield of Camellia oleifera used in this paper,the random forest model has the best estimation effect.Comparing the three estimation models,the coefficient of determination of the random forest model is better than that of the multiple linear regression and the k-nearest neighbor model,except that the single-tree parameter modeling is used alone.Based on the optimal feature set,the R2 of the random forest reaches 0.64.The research results can provide a reference for the use of unmanned aerial vehicle remote sensing technology to carry out Camellia oleifera production surveys in the region. |
