Extraction Of Individual Tree And Stand Structure Parameters Based On RGB Image Of UAV In Closed Canopy Mountainous Chinese Fir Plantation

It is necessary for obtaining the information of all individual trees in the whole stand to carry out precise forest management,but the traditional forest inventorying method cannot satisfy this demand efficiently.The visible light remote sensing technology of UAV not only has obviously advantages in the forest inventorying of designated time and space,but also is easy to be widely used for low cost.However,the studies of the application of UAV visible light image in the extraction of individual tree and stand structure parameters in the closed canopy mountainous forest are still scarce.In this paper,selecting the typical subtropical closed canopy mountainous Chinese fir plantation was as the study object,the research of extracting the individual tree and stand structure parameters was carried out based on the high spatial resolution images taken by the DJI Phantom 4 Pro UAV,and the results were used to the improvement thinning design.The main results are as follows:(1)The individual tree detection method of local maximum based on DSM data has higher detection accuracyA method of local maximum individual tree detection based on DSM data is proposed and tested in this study with the center position of the stump measured by RTK as the true value of the individual tree coordinate.The proposed method is compared with four others commonly used individual tree detection methods(local maximum individual tree detection method based on orthophoto image,individual tree detection method based on rule-based object-oriented classification,individual tree detection method based on multi-scale segmentation and manual visual interpretation).It is found that the overall accuracy of the method proposed in this paper is the highest,reaching 84.5%,and it can keep consistent and relatively low in the error of repeated and wrong detection.(2)Marker-controlled watershed algorithm can effectively extract crown area of individual treeIn this study,the result of individual tree detection of local maximum based on DSM data is used as the foreground marker data,which is overlapped with DOM image,and the marker-controlled watershed segmentation is programmed to rapidly extract the tree crown area.By analyzing the accuracy of the crown extraction results obtained by watershed segmentation and the area of the crown drawn by visual inspection,it is found that there is a linear relationship between these two methods.The R~2 of the regression model of the crown area is 0.9488,and the overall accuracy is 96.31%.(3)The classification accuracy based on VEG vegetation index of fine sub-compartments is higherBased on the spatial crown density heterogeneity and the true value of visual inspection,comparing the accuracy of the fine sub-compartments divided by eight indexes,it is found that VEG vegetation index is the closest to the true value.The overall accuracy is 77.0%,and the Kappa coefficient is 0.679.(4)The fitting DBH-crown model and tree height-crown model have high accuracyThe field measured DBH and tree height and crown width obtained from visible light image of UAV are used to fit the DBH–crown model and height–crown model,respectively.It is found that the optimal fitting models are linear model and composite model.The highest determinant coefficients of both models are 0.759 and 0.688.The accuracy is between(75.62%,99.99%)and(88.75%,99.97%)and the average accuracy is 94.64%and 97.27%,respectively.(5)Different regions in the same sub-compartment have different stand structure characteristicsBased on the results of individual tree detection and crown segmentation,the stand structure parameters of the whole stand area and multiple fine sub-compartments are calculated respectively.It is found that there are significant differences in the stand structure parameters between the continuous fine sub-compartments,which indicates that it is necessary to divide small sub-compartments.(6)Fine sub-compartments can be used to design thinningBy comparing the thinning design based on the structure parameters of the fine sub-compartments with the other two thinning ways(the thinning based on the structure parameters of the whole area and the thinning based on the structure parameters of the whole area),it is found that the thinning intensity of the three ways has significantly difference.The thinning design based on fine sub-compartments can customize the harvesting intensity according to the spatial heterogeneity.In conclusion,this study shows that the UAV visible light remote sensing image using the local maximum method and the marker-controlled watershed algorithm can effectively extract the individual tree and stand structure parameters,which has substantive guiding significance and practical value for the precise forest management.