Research On Single Tree Species Identification Based On UAV Hyperspectral Collaborative LiDAR Data

The accurate classification of forest individual tree species is significant for assessing species diversity and estimating forest productivity.Recently,classifying individual trees in forest quickly and accurately is a hot spot in forestry remote sensing research.This paper takes the mixed forest in Maoershan Experimental Forest Farm of Northeast Forestry University as the research object.Based on UAV LiDAR(light detection and ranging)and UAV hyperspectral data,a variety of classification schemes are designed to explore the effects of different data sources,different classifiers,as well as crown morphological characteristics,on the classification accuracy of tree species on the individual tree.Firstly,based on UAV LiDAR data,this study generates the crown height model(CHM)of the sample plot.Then the marker controlled watershed algorithm is used to segment the single tree crown.After that,each segmented tree crown is matched with each tree crown obtained from the field investigation.According to the results of single tree segmentation,the spectral information and LiDAR echo information in the tree crown are extracted as classification variables,and the random out of bag data(OOB)is used to sort and screen the importance of the variables.Finally,three machine learning models,random forest classifier,support vector machine,and BP neural network,are used to classify the segmented individual trees.The three classification results are compared,and the accuracy is evaluated.This study draws the following conclusions:(1)Comparing the results of different remote sensing source,the classification accuracy of multi-source data result is significantly higher than that of single data source result.Meanwhile,when only one data source is used,the classification accuracy using hyperspectral data is higher than that using LiDAR data.(2)Comparing the classification results of three different classifiers,the classification accuracy of the random forest classifier and support vector machine are similar,which is more than 78%.The classification accuracy of BP neural network is the lowest,which is 75.8%.(3)After adding the crown morphological features extracted by UAV LiDAR into the classifier,the classification accuracy of the three classifiers is slightly improved.The kappa coefficient comparison indicates that the consistency between the classification results and the original data is improved after adding the crown contour variable.It is proved that the addition of crown morphological features has a certain positive significance for classification.