时序Sentinel-2A影像光谱特征的茶园提取应用  

The application of spectral characteristics of time series Sentinel-2A images in tea land extraction

作　　者：赵晓晴 王萍[1] 荆林海 谭炳香[3] 赵鑫 刘德军 ZHAO Xiaoqing;WANG Ping;JING Linhai;TAN Bingxiang;ZHAO Xin;LIU Dejun(College of Geomatics,Shandong University of Science and Technology.Qingdao,Shandong 266510,China;Key Laboratory of Digital Earth,Institute of Remote Sensing and Digital Earth,Chinese Academy of Science,Beijing 100094,China;Institution of Forest Resources Information Technique,Chinese Academy of Forestry,Beijing 100091,China;Zhejiang Zhongke Star Map,Hangzhou 310000,China)

机构地区：[1]山东科技大学测绘科学与工程学院,山东青岛266510 [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,北京100094 [3]中国林业科学研究院资源信息研究所,北京100091 [4]浙江中科星图,杭州310000

出　　处：《测绘科学》2020年第6期80-88,共9页Science of Surveying and Mapping

摘　　要：针对当前茶园遥感识别研究未充分分析利用光谱特征提取茶园的问题,该文详细探讨了光谱时间变化特征在茶园遥感识别中的应用潜力。研究利用时序Sentinel-2A影像,分析7种典型地物的时序光谱变化与NDVI变化,发掘出可用于区分茶园与其他地物的特征波段——红边2、红边3、近红外、红边4、短波红外1、短波红外2。基于上述波段或NDVI构造18个茶园提取特征,最终确定14个茶园提取特征,并基于每种特征分别构建决策树,实现茶园提取,并验证每种特征的可行性。结果表明,分类精度较高的前3个特征分别为SR-SWIR2-NIRMay、SD-NIR-SWIR2May、SR-NDVIMay&Dec;总体精度依次为96.71%、94.24%、93.43%。In view of the current research on remote sensing identification of tea land,the problem of extracting tea land by spectral features was not fully analyzed.,this paper explored the application potential of spectral time variation characteristics in tea land remote sensing recognition.This study mainly used time series Sentinel-2A images to analyse the spectral changes of seven objects to compare the the particularity between tea land and others,and decision tree classification was utilized to realize tea land extraction.By analyzing the time series spectral changes and NDVI changes of seven typical objects,the characteristic bands that could be used to distinguish tea land from other objects were discovered:red-edge 2,red edge-3,near infrared,red-edge 4,short-wave infrared 1,short-wave infrared 2.Based on the above mentioned bands or NDVI,18 tea land extraction indexes were constructed,it was verified that 14 tea land extraction indexes were finally determined and decision trees were constructed respectively to realize tea land extraction based on each feature.The evaluation indicators showed that the top three indexes were:SR-SWIR2-NIR May,SD-NIR-SWIR2_Mny,SR-NDVI.May&Dee,proved the effectiveness of the three extraction features,and the overall accuracy was 96.71%,94.24%and 93.43%.

关 键 词：茶园提取 Sentinel-2A 光谱特征 归一化植被指数 决策树 

分 类 号：P237[天文地球—摄影测量与遥感]