摘要
提出针对Landsat 8影像的云识别方法SARM.在对云及其他地物进行光谱分析的基础上,使用Landsat 8可见光到近红外波段(波段1~5)和热红外波段(波段10、11),构建基于像元的波谱面积比值.利用归一化植被指数(NDVI)和波谱面积比值构建影像的散点图,采用高、中、低3种云识别置信区间,完成对云的识别.以3景不同地区的Landsat 8影像为例进行实验,每景选取具有代表性的3个区域,每个区域10 000个像元进行精度分析.结果表明:波谱面积比值增强了云和下垫面的差异,更利于区分;基于波谱面积比值和NDVI的散点图,能够清晰地展现不同地类条件下云的分布特征;利用可调阈值的提取方法能够满足不同研究目的对云识别的需求;与已提出的3种云识别方法相比,总体精度提高10%左右.
The spectral area ratio method（SARM）was proposed to detect clouds from Landsat 8 images.Landsat 8 visible to near infrared bands（band 1-band 5）and thermal infrared bands（band 10-band 11）were used based on spectral analysis on clouds and other ground objects in order to establish pixels-based spectral area ratio.Scatter plot for images was plotted with normalized difference vegetation index（NDVI）and spectral area ratio.Clouds were detected with different confidence intervals（high,medium and low level）.Three Landsat 8 images of different spatial were employed to demonstrate the accuracy of the method with three representative zones from each image and 10 000 pixels from each zone.Results show that spectral area ratio can enhance the difference between clouds and underlying surface,which is beneficial for the cloud detection.Scatter plot based on NDVI and spectral area ratio can clearly display the cloud distribution features under different ground conditions.The extraction method of adjustable threshold can meet requirements of cloud detection with various objectives.The method can significantly improve the overall accuracy by 10% compared with three previous cloud detection methods.
作者
郭仲皓
任宇鹏
秦怡
王鑫
谷娟
马静宇
邹乐君
沈晓华
GUO Zhong-hao, REN Yu-peng, QIN Yi, WANG Xin, GU Juan, MA Jing-yu,ZOU Le-jun, SHEN Xiao-hua(Research Center for Structures in Oil and Gas Bearing Basins Ministry of Education, Zhejiang University,Hangzhou 310027, China)
出处
《浙江大学学报（工学版）》
EI
CAS
CSCD
北大核心
2018年第7期1423-1430,共8页
Journal of Zhejiang University：Engineering Science
基金
国家科技重大专项资助项目(2017ZX05008-001-006)
作者简介
郭仲皓（1993-）,男,硕士,从事遥感图像云识别方法的研究.orcid.org/0000-0003-2757-6985.E-mail：guozhonghao@zju.edu.cn;通信联系人：邹乐君,男,教授.orcid.org/0000-0003-1408-5968.E-mail：zoulejun2006@zju.edu.cn
