文章摘要
周强1,陈洁2,李玉华1.基于Himawari-8卫星的自适应阈值火点判识算法适用性分析[J].海洋气象学报,2020,40(1):127-133
基于Himawari-8卫星的自适应阈值火点判识算法适用性分析
Applicability analysis of adaptive threshold recognition algorithm of fire spot based on Himawari-8 satellite
投稿时间:2019-12-27  修订日期:2020-01-20
DOI:10.19513/j.cnki.issn2096-3599.2020.01.013
中文关键词: Himawari-8  火点识别  敏感性  阈值
英文关键词: Himawari-8  fire spot recognition  sensitivity  threshold
基金项目:山东省气象局预报员专项 (SDYBY2018-04)
作者单位
周强1,陈洁2,李玉华1 (1. 山东省气象服务中心山东 济南 250031 2. 国家卫星气象中心北京 100081) 
摘要点击次数: 1378
全文下载次数: 450
中文摘要:
   为对比分析Himawari-8(葵花8号)静止气象卫星在不同地区反演火点效果,选取2017—2019年山东区域火点统计资料和卫星多通道监测数据,分析自适应阈值判识算法在山东的火点识别效果,针对算法中背景亮温(T3.9bg),背景窗亮温差(ΔT3.9_11bg),以及背景系数n1和n2进行林、草下垫面参数敏感性试验,并通过选取阈值干预算法检验了识别效果。结果表明,自适应阈值算法可多时次连续观测山东区域的火情监测信息,火点数具有明显季节变化特征,遥感识别准确率达71.5%;不同下垫面类型在四个参数试验中识别准确率变化趋势近似一致,但对阈值变化较为敏感;阈值干预算法对林地和草地的火点识别准确率样本均值为72.7%和81.6%,比原始算法分别提升了5.8%和3.8%;阈值干预识别算法能有效过滤误判像元,但两种算法均有漏判区域,根据本地化的下垫面属性优化阈值设定能够有效提升火点识别准确性。
英文摘要:
   In order to compare and analyze the capability of Himawari-8 geostationary meteorological satellite products in retrieving fire spot in different regions, the statistical data of fire spot in Shandong from 2017 to 2019 and the multi-channel monitoring data of satellite are selected to analyze the effectiveness of adaptive threshold recognition algorithm in Shandong Province. With respect to the brightness temperature of background T3.9bg, brightness temperature differente in background window ΔT3.9_11bg, and the background coefficients of n1 and n2, the parameters sensitivity test of the underlying surface of forest and grass is carried out, and the recognition effectiveness is tested by selecting the threshold intervention algorithm. The results are as follows. 1) Adaptive threshold algorithm can continuously conduct multi-time detection of fire spots in Shandong, the number of which shows obvious seasonal variations, and the recognition accuracy is 71.5%. 2) In the 4 parameter experiments, the trend of recognition accuracy of different underlying surface types is similar, but it is sensitive to the change of threshold value. 3) The average accuracy rates of the threshold intervention algorithm for forest land and grassland fire spot recognition are 72.7% and 81.6%, respectively, which are 5.8% and 3.8% higher than those of the original algorithm. 4) The recognition algorithm of threshold intervention can effectively filter misjudged pixels, but both of them have missed areas. Optimizing threshold setting according to the local underlying surface attributes can effectively improve the accuracy of fire spot recognition.
查看全文   查看/发表评论  下载PDF阅读器
关闭