基于多时相CHRIS高光谱卫星数据的优势树种分类研究  

Dominant Tree Species Mapping Based on Multi-temporal CHRIS Hyperspectral Satellite Data

作　　者：韩文军 张苏 焦全军 吴骅[3] HAN Wenjun;ZHANG Su;JIAO Quanjun;WU Hua(Engineering Data Center,State Grid Economic and TechnologicalResearch Institute Co.,Ltd.,Beijing102209,China;Key Laboratory of Digital Earth Science,Institute of RemoteSensing and Digital Earth,Chinese Academy of Sciences,Beijing100049,China;State Key Laboratory of Resources and Environmental Information System,Institute of Geographic Science and Natural Resource Research,Chinese Academy ofSciences,Beijing100101,China)

机构地区：[1]国网经济技术研究院有限公司工程数据中心,北京102209 [2]中国科学院遥感与数字地球研究所数字地球重点实验室,北京100049 [3]中国科学院地理科学与资源研究所资源与环境信息系统国家重点实验室,北京100101

出　　处：《林业调查规划》2019年第2期1-6,共6页Forest Inventory and Planning

基　　金：国网经研院自主投入科技项目(ZZKJ-2018-10);国家重点研发计划项目(2016YFD0300601);海南省自然科学基金项目(20164180)

摘　　要：多时相高光谱卫星遥感包含树种光谱特征和生长季相差异信息,是解决森林树种识别精度不足的重要技术途径。本文利用不同季节3个时相CHRIS高光谱卫星影像,设计了以Bhattacharyya距离为可分性准则的波段选择经验算法,实现吉林省汪清研究区的优势树种多时相高光谱卫星填图。结果显示,多时相高光谱数据的可分性指标相比单一时相增幅明显;结合波段选择的多时相高光谱分类结果验证精度较单一时相分类结果和多时相全波段分类精度提高7.5%和1.6%;研究区主要优势树种的分类精度存在差异,柞树和落叶松的分类精度最高,杨树最低,红松与暗针叶林存在一定程度的误分,主要原因为二者的光谱接近且时相特征差异小。The multi-temporal hyperspectral satellite remote sensing,which contains spectral characteristics and seasonal difference information of tree species,is an important technical approach to solve the inaccuracy of forest tree species identification. The three-temporal CHRIS hyperspectral satellite images of different seasons were used to design an empirical algorithm for band selection based on the separability criterion of Bhattacharyya distance,so as to realize the multi-temporal hyperspectral satellite mapping of dominant tree species in the study area of Wangqing,Jilin Province. The results showed that the separability index of multi-temporal hyperspectral data increased obviously compared with that of single temporal data. The multi- temporal hyperspectral classification method involving band selection improved the accuracy of tree species identification by 7.5% and1.5% respectively compared with that of single temporal classification and multi-temporal classification with band. The classification accuracy of dominant tree species in the study area was different. The highest classification accuracy was Quercus mongolica and Larix olgensis ,the lowest was Populus canadensis Moench.The similar spectra and temporal characteristics of Pinus koraiensis and dark coniferous forest caused some misclassification.

关 键 词：多时相CHRIS高光谱卫星影像 树种分类 光谱特征 BHATTACHARYYA距离 波段选择 可分性准则 

分 类 号：S79[农业科学—林木遗传育种] P228S711[农业科学—林学]