Airborne hyperspectral data can reflect the spectral characteristics of tree species, which can be used for precise classification of forest tree species. This study applies different machine-learning classification algorithms to classify tree species in hyperspectral images of unmanned aerial vehicles (UAV). Firstly, a UAV was used to collect hyperspectral data from the Maor Mountain Experimental Forest Farm in Heilongjiang Province, and a series of preprocessing was completed for the obtained data. Then, three different machine learning classification algorithms, namely, support vector machine (SVM) based on Gaussian kernel, random forest (RF), and K-nearest neighbor (KNN), were used to establish the tree species classification models, respectively, based on the full-band hyperspectral data. Meanwhile, tree species classification models were constructed based on the dimension-reduced full-band hyperspectral data using different band selection methods (successive projections algorithm, competitive adaptive reweighted sampling method and uninformative variable elimination method). Finally, the tree species classification model was constructed by combining different band selection methods and hyperspectral image texture features, and the results of different processing methods were compared. Research shows, the kernel SVM had the highest classification accuracy (87.55%) among the tree species classification models with full-band hyperspectral data. After selecting different bands, the stability of RF is the best among the three classification algorithms, and the accuracy rate was high, while the classification accuracy of the support vector machine based on the Gaussian kernel improved with the increase of feature dimension. The accuracy of the tree species classification model established by extracting texture features based on a grayscale co-occurrence matrix combined with band selection was higher than that of the model established by a single band selection. In particular, the K-nearest neighbor classification algorithm has the greatest improvement, which indicated that modeling with clearly partitioned features can achieve good classification results. This study used different feature selection methods combined with three different machine learning classification algorithms to achieve dominant tree species classification based on hyperspectral data, which provides technical reference for the combination of band selection methods and machine learning algorithms, and it is also of great significance for forest biomass retrieval and carbon storage estimation based on UAV hyperspectral data.
经过SPA波段选择之后共筛选出2个特征波段。图3为利用SPA提取树种高光谱数据全部特征波段的运算结果。对300个波段变量建立了偏最小二乘回归提取波长后的多元线性回归模型,由图3(a)可以看出,随着采样次数增加,波段数由最初的300个,最终选择波段个数为2个,此时的均方根误差(Root mean squared error,RMSE)值最小,RMSE达到最小值1.208,此时的模型预测精度已达到最高,由图3(b)可以看出,选出来的波段变量是第166和174波段。
由图4可以看出,随着蒙特卡罗迭代次数的增加,被选择的波长数量不断下降,当蒙特卡罗迭代次数为18时,交叉验证均方差(Root mean square error of cross validation,RMSECV) 的值最小,此时自适应加权采样选择的特征波长建立的偏最小二乘回归的预测效果最好,选择出的48个波段变量为最优变量组合。
ZHONGH, LIUH R, LINW S.Application of lidar and hyperspectral remote sensing technology to tree species identification[J].World Forestry Research,2021,34(4):41-45.
[3]
吴倩.基于机载高光谱遥感数据的森林乔木树种多样性研究[D].合肥:安徽农业大学,2018.
[4]
WUQ.Research on vegetation diversity of forest Arbor species based on airborne hyperspectral remote sensing data[D].Hefei:Anhui Agricultural University,2018
[5]
KOUKALT, ATZBERGERC, SCHNEIDERW.Evaluation of semi-empirical BRDF models inverted against multi-angle data from a digital airborne frame camera for enhancing forest type classification[J].Remote Sensing of Environment,2014(151):27-43.
[6]
DALPONTEM, ØRKAH O, GOBAKKENT,et al.Tree species classification in boreal forests with hyperspectral data[J].IEEE Transactions on Geoscience and Remote Sensing,2021,51(5):2632-2645.
ZHAOQ Z, JIANGP, WANGX W,et al.Classification of protection forest tree species based on UAV hyperspectral data[J].Transactions of the Chinese Society for Agricultural Machinery,2021,52(11):190-199.
[9]
BALLANTIL, BLESIUSL, HINESE,et al.Tree species classification using hyperspectral imagery:a comparison of two classifiers[J].Remote Sensing,2016,8(6):445.
FANX, LIUQ W, TANB X.Classification of forest species using airborne PHI hyperspectral data[J].Remote Sensing for Land and Resources,2017,29(2):110-116
MIAOY H, YANGM, WUG J.Sophisticated vegetation classification based on multi-dimensional features of hyperspectral image[J].Journal of Atmospheric and Environmental Optics,2020,15(2):117-124.
[14]
ZHONGH, LINW S, LIUH R,et al.Identification of tree species based on the fusion of UAV hyperspectral image and LiDAR data in a coniferous and broad-leaved mixed forest in Northeast China[J].Frontier in Plant Science,2022,13:964769.
[15]
陆嘉辉.基于林冠高光谱影像的树种分类研究[D].上海:华东师范大学,2020.
[16]
LUJ H.Tree species classification based on canopy hyperspectral images[D].Shanghai:East China Normal University,2020.
[17]
LIUC, AIM, CHENZ,et al.Detection of Firmiana danxiaensis canopies by a customized imaging system mounted on an UAV platform[J].Journal of Sensors,2018(9):1-12
CHENY, DONGL B, LIUZ G.Optimal tree species composition of main stand types in Maoershan natural secondary forest[J].Journal of Beijing Forestry University,2019,41(5):118-126.
[20]
林海军.塔里木河下游主要荒漠植物高光谱特征分析[D].乌鲁木齐:新疆农业大学,2014.
[21]
LINH J.Analysis of spectral features of main desert plants in the lower reaches of Tarim River[D].Urumqi:Xinjiang Agricultural University,2014.
LIUM B, TANGY L, LIX L,et al.Feasibility of using successive projections algorithm in spectral monitoring of rice leaves nitrogen contents[J].Infrared and Laser Engineering,2014,43(4):1265-1271.
ZHOUH P, HUY L, JIANGH Z,et al.Detection method of oil content of Camellia oleifera seed based on hyperspectral imaging[J].Transactions of the Chinese Society for Agricultural Machinery,2021,52(5):308-315.
CHENY Y, WANGZ B, WANGZ B.Novel variable selection method based on uninformative variable elimination and ridge extreme learning machine:CO gas concentration retrieval trial[J].Spectroscopy and Spectral Analysis,2017,37(1):299-305.
[30]
李丽.基于ELM的高光谱遥感影像土地利用覆盖分类优化方法研究[D].成都:电子科技大学,2018.
[31]
LIL.Research on land use cover classification optimization methods for hyperspectral remote sensing image based on ELM[D].Chengdu:University of Electronic Science and Technology of China,2018.
KONGJ X.Object-based tree species classification and identification using Unmanned Aerial Vehicle remote sensing imageries in a subtropical evergreen deciduous broad-leaved mixed forest[D].Shanghai:East China Normal University,2020.
[34]
臧卓.南方主要乔木树种高光谱数据降维组合分类算法研究[D].长沙:中南大学,2015.
[35]
ZANGZ.Combination and comparison of dimensional reduction and classification algorithms for classifying main tree species of Southern China base on hyperspectral data[D].Changsha:Central South University,2015.
[36]
江萍.基于无人机高光谱影像的防护林树种分类研究[D].石河子:石河子大学,2022.
[37]
JIANGP.Research on tree species classification of shelterbelt based on UAV hyperspectral image[D].Shihezi:Shihezi University,2022.
LIUY S, LÜC W, ZHUF X,et al.Extraction of high spatial resolution remote sensing image classification based on PCA and multi-scale texture feature[J].Remote Sensing Technology and Application,2012,27(5):706-711.
YUANJ X.Study on hyperspectral remote sensing identification method of dominant mangrove species——taking a typical area in Guangdong Province as an example[D].Qingdao:China University of Petroleum(East China),2021.
CUIL L.Integrative analysis and evaluation of the interpretation features in remote sensing image[D].Beijing:Graduate School of Chinese Academy of Sciences(Institute of Remote Sensing Applications),2005.
JIANGY F, QIJ G, CHENB W,et al.Classification of mangrove species with UAV hyperspectral imagery and machine learning methods[J].Remote Sensing Technology and Application,2021,36(6):1416-1424.
YUH, TANB X, SHENM T,et al.Research on identification of dominant tree species using airborne hyperspectral images based on machine learning algorithm[J].Remote Sensing for Natural Resources:1-10[2023-12-11].
[52]
徐新良,曹明奎.森林生物量遥感估算与应用分析[J].地球信息科学,2006(4):122-128.
[53]
XUX L, CAOM K.An analysis of the applications of remote sensing method to the forest biomass estimation[J].Geo-information Science,2006(4):122-128.
[54]
李军玲.机载高光谱影像温带人工林树种分类[D].北京:中国林业科学研究院,2019.
[55]
LIJ L.Species classification of temperate plantation using airborne hyperspectral images[D].Beijing:Chinese Academy of Forestry,2019.