基于无人机高光谱数据和3D-2D-CNN的天然次生林主要树种分类研究

打开文本图片集

关键词：天然次生林;树种分类;高光谱图像;3D-2D-CNN模型;卷积神经网络;机器学习;波段重要性分析中图分类号：S757 文献标志码：A 文章编号：1000-2006（2026）02-0009-10

Discriminate dominant tree species in natural secondary forests from UAV hyperspectral images using a hybrid 3D-2D convolutional neural network

LI Hao，QUAN Ying，LIU Jianyang，BIAN Shaojie，WANG Bin，LI Mingze （SchoolofForestry，NortheastForestryUniversity，Harbin15OO4O，China）

Abstract：【Objective】Thisstudyaims to improvethe clasificationaccuracyof dominant tree species in typical natural secondary forests in northeast China，aconvolutional neural network（CNN）-based frameworkfortreespecies classification using UAV hyperspectral images was proposed.【Method】Four dominant tree species—Fraxinus mandshurica，Juglans mandshurica，Ulmus sp.，and Betula platyphylla—from the Maoershan Experimental ForestFarm of NortheastForestry Universitywere studied.Hyperspectral images ofseven diferentregions were acquiredusinganovel UAV-mounted hyperspectral imager.Asingle-tree dataset with varying crownsizeswas constructedusinggroundmeasured data，divided into training and test sets at a 7：3 ratio.A hybrid 3D-2D-CNN model integrating 3D and 2D convolutionallayerswasdeveloped：3Dconvolutionallayersextractedspectral-spatialcoupledfeatures，while2Dlayers captureddetailed spatial features，enhancingthe model'sholistic learningcapability.The model was compared with 2DCNN，3D-CNN，and feature-selection-based machine learning models （random forest （RF），support vector machine （SVM），and gradient boosting machine （GBM））. Additionally，the band importance was analyzed using a progressive bandremoval method，and spectral feature sensitivitywas investigated.【Result】The proposed 3D-2D-CNNmodel achieved a classification accuracyof 87% and an F1 score of 0.86 for the four tree species，outperforming other algorithms with an overall accuracy improvement of 5%-6% .Band importance analysis highlighted the significant contributionof thenear-infraredbandclassification.【Conclusion】The3D-2D-CNNmodel，byefectivelyintegrating spectralandspatial information，significantlyenhancedtheclasificationperformanceofnatural secondaryforesttree species compared to traditional methods.This approach provides technical support for forest resource management and ecosystem monitoring via remote sensing.

Keywords：natural secondary forest; treespecies classification; hyperspectral image；3D-2D-CNNmodel；convolutional neural network；machine learning；band importance analysis

随着机载遥感技术的快速发展，高空间、高光谱分辨率的影像已成为获取森林细致结构和树种信息的重要工具[1-2]。（剩余21465字）