森林工程2026，Vol.42Issue(1)：11-22,12.DOI:10.7525/j.issn.1006-8023.2026.01.002
联合无人机高光谱与激光雷达数据的森林树种识别
Tree Species Identification by Combined of UAV Hyperspectral and LiDAR Data
摘要
Abstract
Identifying the types and distribution of tree species is the foundation for monitoring tree diversity,and is cru-cial for forest protection and management and sustainable development of forests.Forest plot-scale hyperspectral images and LiDAR point cloud data scanned by unmanned aerial vehicle(UAV)were used as the data source,and based on the individual tree-scale hyperspectral and point cloud data obtained by the canopy height model,a convolutional neural net-work(CNN-EGNet)model combined with the attention mechanism of efficient channel attention(ECA)was proposed in this study,aiming at achieve precise tree species identification in mixed coniferous and broadleaf forests in the Maoer-shan area of Shangzhi City,Then,CNN-EGNet with three traditional CNN models VGG16,VGG19,and GoogLeNet in identification accuracy.Finally,on the basis of the results of the tree species identification,tree species diversity indi-ces(Shannon-Wiener,Simpson,Pielou,Species richness)in the study area were calculated with the 40 m×40 m win-dow.Results showed that the proposed CNN-EGNet model achieved an overall accuracy(OA)of 89.58%and a Kappa coefficient of 0.8661.Compared with conventional models,the overall accuracy for species identification improved by 9.37%,5.20%,and 14.58%,and the Kappa coefficients increased by 0.1175,0.0652,and 0.1896,respectively.The Shannon-Wiener index primarily ranged from 0.8 to 1.4,while the Simpson index predominantly clustered between 0.5 and 0.7.The Pielou index generally fell within the range of 0.7 to 0.95,and the species richness index mostly var-ied between 3 and 5 species.The tree species diversity indices indicated that the distribution of tree species was uneven,with certain species being dominant while others were relatively scarce.The results of the study can provide technical and data references for the identification of tree species and the protection and management of tree species diversity in the mixed coniferous and broadleaf forests in the Northeast of China,and validate the possibility of identification,moni-toring and evaluating the diversity of tree species by combining hyperspectral and LiDAR data from UAVs with convolu-tional neural networks.关键词
卷积神经网络/高光谱/激光雷达/森林树种多样性/无人机Key words
Convolutional neural network/hyperspectral/LiDAR/tree species diversity/UAV分类
农业科技引用本文复制引用
褚昱凯,林文树..联合无人机高光谱与激光雷达数据的森林树种识别[J].森林工程,2026,42(1):11-22,12.基金项目
黑龙江省自然科学基金联合引导项目(LH2020C049). （LH2020C049）
