基于朴素贝叶斯方法的FY-4A/AGRI云检测模型  

An FY-4A/AGRI cloud detection model based on the naive Bayes algorithm

作　　者：鄢俊洁 郭雪星 瞿建华 韩旻 YAN Junjie;GUO Xuexing;QU Jianhua;HAN Min(Beijing Huayun Shinetek Science and Technology Co.,Ltd.,Beijing 100081,China)

机构地区：[1]北京华云星地通科技有限公司,北京100081

出　　处：《自然资源遥感》2022年第3期33-42,共10页Remote Sensing for Natural Resources

基　　金：国家自然科学基金面上项目“基于深对流云和月球高低端目标的长序列气象卫星辐射定标研究”(编号:41675036)资助。

摘　　要：针对风云四号A星(FY-4A)中多通道扫描成像辐射计(advanced geosynchronous radiation imager,AGRI)云检测问题,提出了一种基于朴素贝叶斯算法的全自动云检测方法。使用朴素贝叶斯算法作为核心结构,基于光学载荷基本云检测原理选择合适的红外通道作为特性分类器参数,可保证日夜云检测一致性,同时针对不同的地表类型和不同月份分别分类训练构建,最终得到基于朴素贝叶斯算法的云检测模型。针对FY-4A/AGRI数据生成了7种经典的云检测特征和1种基于红外合成图像特征的贝叶斯分类器,经过2019年国家卫星气象中心业务云检测产品的学习测试验证,在陆地、沙漠、浅水和深海的召回率(probability of detection,POD)达到98%以上,积雪POD达到80%,南北极POD达到80%以上。将检测结果与国家卫星气象中心业务系统云检测结果进行比较,全年月度平均POD均高于98%,误判率(false alarm ratio,FAR)低于5%,Kuipers评分(Kuiper’s skill score,KSS)均高于90%。This study developed an automatic cloud detection method based on the naive Bayes algorithm for the cloud detection of the advanced geosynchronous radiation imager(AGRI)aboard the FY-4A satellite.In this method,the naive Bayes algorithm serves as the core structure,and appropriate infrared channels are selected as the parameters of the characteristic classifier according to the basic cloud detection principle of optical payload to ensure the consistency of cloud detection between day and night.After the classified training and construction for different surface types and different months,a cloud detection model based on the naive Bayes algorithm was finally established.Moreover,the classifier for FY-4A/AGRI data used in the method was established considering seven typical cloud detection characteristics and one characteristic based on the infrared composite images.As indicated by the learning tests and verification using the business cloud detection product of the National Satellite Meteorological Center(NSMC)in 2019,the classifier yielded a probability of detection(POD)greater than 98%for land,desert,shallow water,and deep sea,greater than 80%for snow cover,and greater than 80%for North and South poles.The comparison between the cloud detection results of this study and those obtained using the NSMC business system showed that the cloud detection results of this study had an average monthly POD of the whole year greater than 98%,a false alarm ratio(FAR)less than 5%,and all Kuiper’s skill scores(KSSs)greater than 90%.

关 键 词：FY-4A/AGRI 朴素贝叶斯 云检测 POD 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]