影像选择及阈值对dNBR提取火烧迹地的影响  

Influence of Image Selection and Threshold on dNBR Extraction of Burned Area

作　　者：韩丽 戴必辉 王秋华 高仲亮[1,2] HAN Li;DAI Bihui;WANG Qiuhua;GAO Zhongliang(College of Civil Engineering,Southwest Forestry University,Kunming 650224,China;Yunnan Key Laboratory of Forest Disaster Warning and Control,Kunming 650224,China)

机构地区：[1]西南林业大学土木工程学院,昆明650224 [2]云南省森林灾害预警与控制重点实验室,昆明650224

出　　处：《遥感信息》2023年第1期47-55,共9页Remote Sensing Information

基　　金：国家自然科学基金项目(31960318、32160376、31860214);云南省教育厅基金项目(2021J0160);云南省农业联合面上项目(2018FG001-055)。

摘　　要：针对缺乏必要的影像选择及阈值评估导致差分归一化燃烧指数(differenced normalized burn ratio,dNBR)提取火烧迹地的可靠性降低问题,文章探索其影响规律。筛选滇中地区15场有高清影像的火灾为研究对象,采用dNBR提取不同时间及阈值组合下的火烧迹地面积,以10场火灾作为研究区确定最佳影像对及阈值,并以5场火灾作为验证。结果表明,影像选择过程及阈值对评估结果影响很大,不合理的影像选择会产生错误的结论,其中灾后影像的影响占主导地位。灾后最优选择是NBR最低值及其附近影像,在0.2~1.2阈值区间提取精度平均为94%,火灾细节勾绘准确。In view of the lack of necessary image selection and threshold evaluation,the reliability of dNBR(differenced normalized burn ratio)extraction is reduced,and its influence law is explored.Select 15 fires with high-definition images in central Yunnan as the research object,take 10 fires as the research area,the best image pair and threshold are determined,then validate the results with other five fires.The results show that the image selection process and threshold have a huge impact on the evaluation results,and unreasonable image selection may even lead to wrong conclusions.Among them,the influence of post-disaster images is dominant.The best choice after the disaster is the image with the lowest NBR value and its vicinity.The average extraction accuracy is 94%in the range of 0.2~1.2 threshold,and the fire details are accurately delineated.

关 键 词：影像选择 归一化燃烧指数 火烧迹地 林火 滇中地区 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]