Individual Tree Extraction And Species Identification Based On Airborne LiDAR And Hyperspectral Data

As an important part of terrestrial ecosystems,forests have a profound impact on the global climate,ecology,and carbon cycle.The forest spatial structure is composed of individual trees.For the investigation of forest resources status and production practices,it is of great significance to accurately and efficiently obtain the single-wood structure parameters.In addition,the identification of forest species can also provide the basis and important basis for the inventory of forest resources,the drawing of forest maps,and the estimation of carbon reserves on forest land.Traditional single-free structure parameters extraction and tree species identification rely mainly on field surveys.With the aid of special instruments,it takes a lot of time,manpower and financial resources.In particular,surveys of large-scale forest resources can only use sample surveys.Hyperspectral remote sensing images contain near-continuous spectral information of the ground objects,and can accurately detect various types of ground objects with subtle spectral differences.It has great advantages in tree species recognition.At the same time,the Light Detection And Ranging(LiDAR)is a new type of remote sensing technology,which has a powerful ability to observe the earth,and has great potential in detecting the spatial structure characteristics of the ground objects.This paper selects the Daxing'an Lingen River Forest Reserve as the research area,and takes the airborne laser radar data and its simultaneous acquisition of the hyperspectral remote sensing image as the data source to carry out single tree parameter extraction and tree species classification research.Firstly,LiDAR point cloud data is divided into ground points and non-ground points by a triangulated gradient progressive filtering algorithm,and inverse distance weighted interpolation(IDW)is used to generate digital elevation model(DEM).Lidar point cloud is firstly interpolated to generate a Digital Surface Model(DSM).Both of the grids compute a Canopy Height Model(CHM),and the original point cloud data is normalized by DEM..A single wood extraction study was conducted on the generated CHM and normalized point cloud data,and on this basis single tree scale tree species identification was performed.The main conclusions are as follows:(1)In the 100m x 100m large sample plot,355 single trees can be extracted based on the canopy height model single-wood extraction method,in which 322 trees are correctly matched with the actual ground measurements,and the correct rate of separation results is 90.7%;The single wood extraction method based on the normalized point cloud can be used to separate only a small number of single trees,and a total of 221 single trees were extracted,of which 206 were correctly matched with the actual measurements on the ground.The accuracy of the separation results was 93.21%.(2)The tree height parameters obtained by single tree extraction method based on the canopy height model are 0.71 and the average precision is 81.2%.The tree height parameters obtained by the single tree extraction method based on the normalized point cloud are 0.72.The R2 measured with the ground was 0.73 with an average accuracy of 82.6%.The crown amplitude data extracted by the two methods are quite different from the measured crown amplitudes.The crown width and the actual measurement crown R2 of the former are 0.42.The crown width and the actual crown radius R2 extracted by the latter are only 0.21.Crown amplitude is much larger than the measured crown.(3)Based on the results of individual trees based on canopy height model extracted from single trees,the hyperspectral data were used to extract the single tree spectral features,and the classification of larch and birch trees in the plot was conducted using a random forest classifier.The overall accuracy of the classification is 91.61%,and the Kappa coefficient is 0.7819.