基于Sentinel-2光谱与地形特征的山区森林分类——以武夷山国家公园为例  

Mountain forests classification using the Sentinel-2 spectral features and topographic characteristics:a case study of wuyishan national park

作　　者：张春莹 江洪[1,2,3] 林敬兰 岳辉 Zhang Chunying;Jiang Hong;Lin Jinglan;Yue Hui(Key Lab of Spatial Data Mining and Information Sharing of Ministry of Education,National&Local Joint Engineering Research Center of Satellite Geospatial Information Technology,Academy of Digital China,Fuzhou University,Fuzhou 350108,China;Fujian Soil and Water Conservation Experiment Station,Fuzhou 350002,China;Soil and Water Conservation Center of Changting County,Changting 366300,China)

机构地区：[1]福州大学空间数据挖掘和信息共享教育部重点实验室,卫星空间信息技术综合应用国家地方联合工程研究中心,数字中国研究院(福建),福建福州350108 [2]福建省水土保持试验站,福建福州350002 [3]长汀县水土保持中心,福建长汀366300

出　　处：《海南大学学报（自然科学版）》2023年第2期160-171,共12页Natural Science Journal of Hainan University

基　　金：福建省科技计划引导性项目(2021Y0005);福建省水利科技项目(MSK202301)。

摘　　要：为了提高山区复杂地形条件下的森林分类精度,以武夷山国家公园为研究区开展山区森林分类研究.以Sentinel-2遥感影像作为数据源,采用多尺度分割方法进行影像分割,并通过ESP2算法选取影像的最优分割尺度.首先构造了Sentinel-2光学影像的初始27维光谱特征,通过计算随机森林Gini指标对分类特征变量进行优化,最终确定17维最优特征变量,然后将提取的研究区地形特征与最优特征变量相结合,应用随机森林算法进行森林分类.结果表明:在27维光谱特征中选取重要性排名前17的特征参与分类时,精度最高值为0.911 0,其中,Sentinel-2影像的红、红边和近红外波段及其相应的光谱指数在森林分类中有较高重要性;在不同的特征参与分类时,在参考光谱特征中依次加入红边指数和地形因子,分类的总体精度分别为88.13%、89.50%、90.87%,Kappa系数分别为0.854 6、0.871 0、0.887 8.研究证明将Sentinel-2丰富的光谱特征与地形因子相结合,可有效获取各森林地物类型在不同地形特征下的不同光谱特征,此方法在森林地物信息提取中具有较高的应用价值,为今后地形复杂的山区森林进行快速、准确的分类提供技术方法参考.In order to improve the accuracy of forest classification under complex terrain conditions in mountainous areas,in the report,Wuyishan National Park was used as the special area to study the forest classification in mountainous areas.Sentinel-2 remote sensing images were used as the data source,a multiscale segmentation method was used for image segmentation,and the ESP2(Estimation Scale Parameter) algorithm was used to select the optimal segmentation scale of the images.The 27 spectral feature variables of the Sentinel-2 optical image were constructed,and the Random Forest Gini index was calculated to optimize the classification feature variables and determine the 17 optimal feature variables.The extracted topographic features of the area were then combined with the optimal feature variables,and the random forest algorithm was used for forest classification.The results showed that the highest accuracy value of 0.911 0 is obtained when the top 17 features in importance are selected to participate in the classification among the 27 dimensional spectral features,among which the red,red-edge,and near-infrared bands of Sentinel-2 images and their corresponding spectral indices have higher importance in the forest classification;when the different features are involved in the classification,the red-edge index and topographic factor are added into the reference spectral features in turn,the overall accuracy of the classification is 88.13%,89.50%,and 90.87%,respectively,with Kappa coefficients of 0.854 6,0.871 0 and 0.887 8,respectively.The method has a high application value in the extraction of forest feature information,and provides a technical reference for rapid and accurate classification of mountain forests with complex topography in the future.

关 键 词：山区地形 森林分类 面向对象 Sentinel-2 红边指数 

分 类 号：TP753[自动化与计算机技术—检测技术与自动化装置]