基于无人机高光谱遥感的太行山经济林树种识别研究  

Study on tree species identification of economic forest in Taihang Mountain based on UAV hyperspectral remote sensing

作　　者：孙一丹 杨晓楠 张海涛 张爱军[1,2] 庞立欣 郭艳超 郭雪涛[6] 梁欣 SUN Yidan;YANG Xiaonan;ZHANG Haitao;ZHANG Aijun;PANG Lixin;GUO Yanchao;GUO Xuetao;LIANG Xin(College of Resources and Environmental Sciences,Hebei Agricultural University,Baoding 071000,China;Hebei Provincial Institute of Mountains Research,Baoding 071000,China;College of Horticulture,Hebei Agricultural University,Baoding 071000,China;Academy of Science and Technology,Hebei Agricultural University,Baoding 071000,China;Binhai Agricultural Research Institute,Hebei Academy of Agriculture and Forestry Sciences,Tangshan 063200,China;Baoding Branch of Central Agricultural Broadcasting School,Baoding 071000,China)

机构地区：[1]河北农业大学资源与环境科学学院,河北保定071000 [2]河北省山区研究所,河北保定071000 [3]河北农业大学园艺学院,河北保定071000 [4]河北农业大学科学技术研究院,河北保定071000 [5]河北省农林科学院滨海农业研究所,河北唐山063200 [6]中央农业广播学校保定分校,河北保定071000

出　　处：《林业与生态科学》2024年第2期123-133,共11页Forestry and Ecological Sciences

基　　金：河北省重点研发计划项目(20324202D;20325001D)资助。

摘　　要：以太行山区经济林种植区为研究对象,通过无人机高光谱遥感数据,构建不同经济林树种高光谱特征数据库,利用CART决策树、最大似然法(Maximum likelihood classifier,MLC)、随机森林(Random forest,RF)和支持向量机(Support vector machine,SVM)等方法,获得高光谱遥感经济林树种最优识别模型。研究结果表明:(1)苹果、杏、柿、樱桃、核桃的反射峰在550 nm、750~950 nm及960 nm附近的水汽吸收带差异明显;(2)简单比值指数(SR)、类胡萝卜素反射指数2(CRI2)、绿波段指数(GRVI)等7种植被指数重要性评分大于0.05,利于经济林树种识别;(3)基于光谱特征波段、植被指数、纹理特征的组合方式通过SVM的分类效果最好,优于MLC和RF算法,总体精度(Overall accuracy,OA)达到95.11%,Kappa系数为0.9158。综上所述,基于特征波段、植被指数、纹理特征3种特征组合并采用支持向量机(SVM)分类的识别方法,为6种树种识别的最佳识别方法。The economic forest planting area in Taihang Mountain Area was taken as the research object.Based on the hyperspectral remote sensing data of unmanned aerial vehicle,the hyperspectral characteristic database of different economic forest species was constructed,and the optimal identification model of economic forest species by hyperspectral remote sensing was obtained by using CART decision tree,maximum likelihood classifier(MLC),random forest(RF)and support vector machine(SVM).The results showed that:(1)The water vapor absorption bands of apple,apricot,persimmon,cherry and walnut were obviously different around the reflection peak of 550 nm,between 750950 nm and around 960 nm;(2)The simple ratio index(SR),carotenoid reflex index 2(CRI2),green band index(GRVI)and other 7 plants were more than 0.05,which were beneficial to the identification of economic forest species;(3)SVM was the best classification method based on spectral characteristic band,vegetation index and texture feature,which was better than MLC and RF algorithm.The overall accuracy(OA)was 95.11%and Kappa coefficient was 0.9158.To sum up,based on the combination of characteristic band,vegetation index and texture features,the identification method of support vector machine(SVM)classification was the best identification method for six tree species.

关 键 词：经济林 树种识别 无人机 高光谱 支持向量机 

分 类 号：S127[农业科学—农业基础科学]