Tree Species Classification Using Airborne Hyperspectral Images And LiDAR Data

The identification of tree species is one of the most basic and key indicators in forest resource monitoring.It has great significance in the actual forest resource survey.This paper considering the advantages and characteristics of airborne hyperspectral imagery and LiDAR data,combined the two data in tree species classification,and evaluated the ability of the two data sources in tree species fine identification at complex forest types.We selected a region with a rich variety of tree species in Gaofeng Jiepai Forest Farm in Guangxi province as the study area,combined airborne hyperspectral imagery and simultaneously acquired LiDAR data to classify tree species.Firstly,segmented image in multi-scale,and used ESP(Estimation of Scale Parameter)segmentation evaluation tool to obtain the optimal segmentation scale,and determined multi-level optimal segmentation parameters by combining actual features of each object.Second,extracted feature variables from airborne hyperspectral data and LiDAR data,ICA(Independent component analysis)transformation feature variables and vegetation indices were extracted from airborne hyperspectral data,and constructed new indices according to spectral characteristics of tree species,extracted texture features by gray level co-occurrence matrix,and extracted height information CHM(canopy height model)from LiDAR data,and then constructed a set with multi-feature variables.Combined object-oriented and stratified classification,which was carried out under optimal segmentation scale,then distinguished non-forest land,other forest land and forest land,and finely classify tree species in forest land.Finally,we selected different feature combinations and used support vector machine,k-nearest neighbor and decision tree to classify tree species,compared the effects of different feature combinations and classifiers on tree species classification accuracy.The results show that multi-level segmentation can better distinguish boundaries of different objects,and the segmentation effect is most consistent with actual contour,which can improve the classification accuracy.Combination of object-oriented and stratified classification can effectively avoid the phenomenon of "salt and pepper" and the mixing of tree species and other objects.Among three classifiers,support vector machine classifier has the best classification effect,with the highest classification accuracy of 94.20%and a Kappa coefficient of 0.9307.The use of random forests can eliminate redundant features and further improve performance and efficiency of classifier.In the selected feature subsets,the newly constructed slope spectral index SL2 is a preferred feature,which increases the separability between tree species.Texture features and height information have a great effect on improving tree species classification accuracy,and can effectively distinguish tree species with similar spectral features,which reduces the interference of "different objects having the same spectrum and the same objects having different spectrum".In general,the application of multi-dimensional features can improve the classification accuracy,and the proposed methods of combining two data sources are effective for fine identification of tree species at complex heterogeneous forest stand in south China,which meets the precision requirements of tree species classification.