Forest Structure Parameter Extraction Based On UAV Aerial Survey Data

Forest plays a vital role in regulating climate,conserving water and soil,maintaining environmental humidity,promoting biodiversity and healthy development of ecosystem,and the height,crown width and DBH of trees are the most basic structural parameters.Traditional field measurement mainly relies on manual measurement,which not only consumes a lot of manpower and material resources,but also has strong subjectivity,large measurement error and long operation cycle.Accurate measurement of basic parameters of individual trees is a key challenge of current forest resource management.With the progress of The Times,Lidar,UAV and other advanced remote sensing technologies have begun to appear in our field of vision.The strong penetration of Lidar combined with the unique operating mode of UAV platform from above has opened up a new way for forestry investigation.It is difficult to obtain high-precision forest parameter information from a single data source.In this paper,the UAV Lidar point cloud data with three-dimensional information is combined with ortho image data with high spatial resolution,which can not only obtain highresolution images including texture and spectrum,but also obtain three-dimensional information of single wood and terrain.Canopy height and canopy width information of 35 coniferous forest plots in Gannan Plateau forest were more accurately extracted,diameters at breast height were inverted,and the extraction accuracy was verified with measured data.The main work and conclusions are as follows:(1)For field survey,the UAV is equipped with the laser radar system to collect point cloud data,and the UAV is equipped with a professional camera to acquire ortho image data,and the data is preprocessed by Pixel4 D software.(2)Based on the 3D point cloud data of the sample plot,the digital surface model and digital elevation model were generated by separating vegetation points and ground points with lidar360 software.The three forest canopy heights of low canopy closure,medium canopy closure and high canopy closure were extracted by single tree segmentation based on canopy height model and single tree segmentation based on point cloud,and the accuracy was evaluated by combining the field measured data.The results showed that the segmentation accuracy of single wood based on CHM was 83.9%,R2 was 0.92,and RMSE was 2.55.The accuracy of single tree segmentation based on point cloud was 61.36%,R2 was 0.64,and RMSE was 3.76.That is,the effect of CHM single tree segmentation is better than that of point cloud segmentation and the difference is significant.However,when the two methods extracted the forest canopy heights of different canopy densities,the effect of low canopy densities was better than that of medium canopy densities.This is because the airborne lidar scanning feature from top to bottom can better identify the top of trees.The environment under high canopy densities is complex,so few laser points can hit the ground during flight,and there is a phenomenon that small trees are blocked by big trees in dense forests.The complete tree top cannot be obtained,so the error of high canopy density is large.(3)e Cognition software was used to extract object-oriented crown width information based on two-dimensional orthophoto of sample site.ESP tool was used to screen the optimal segmentation scale for multi-scale segmentation.On this basis,membership function or nearest neighbor classification method was selected to classify crown width and gap according to the actual situation of the sample site,and crown width was extracted from Arc GIS.In the classification of canopy width and gap by object-oriented extraction,the classification progress of canopy width and gap of low canopy closure was slightly higher than that of medium and high canopy closure,and the classification accuracy of middle and high canopy closure was roughly the same,and the overall classification accuracy reached92.78%.Kappa coefficient was 0.86,indicating a good classification effect.However,misclassification still exists.This is because the shadows generated by light in non-crown areas are identified as crown objects in the processing process,resulting in reduced crown extraction accuracy.(4)Pearson correlation analysis was performed on tree height,crown width and DBH,and six unary DBH inversion models including linear,logarithmic,exponential,power function,quadratic polynomial and cubic polynomial were selected.The results showed that the linear model of low canopy closure DBH-tree height model had the best fitting effect.The fitting effect of quadratic polynomial in low canopy closure DBH-crown width model is the best.In the middle canopy closure DBH-tree height model and DBH-crown model,the quadratic polynomial fit is the best.In high canopy closure,both DBH-tree height model and DBH-crown model have the highest fitting degree.In the binary DBH inversion model,the linear model is the optimal model for the three types of canopy density,and the R2 is more than 90%.It can be seen that the DBH model can be used to estimate the DBH well,and the DBH estimation model based on the cooperation of two factors has higher accuracy than that of the single DBH model.