主成分分析与BP神经网络结合的叶绿素a浓度反演模型  

Study on chlorophyll a concentration inversion method based on principal component analysis and BP neural network

作　　者：解学通[1] 郑艳 张金兰[2] 陈克海 XIE Xuetong;ZHENG Yan;ZHANG Jinlan;CHEN Kehai(School of Geography and Remote Sensing,Guangzhou University,Guangzhou 510006,China;School of Mapping and Remote Sensing,Guangdong Polytechnic of Industry and Commerce,Guangzhou 510510,China)

机构地区：[1]广州大学地理科学与遥感学院,广州510006 [2]广东工贸职业技术学院测绘遥感信息学院,广州510510

出　　处：《测绘科学》2024年第10期50-58,共9页Science of Surveying and Mapping

基　　金：广东省海洋经济发展专项项目(GDNRC[2020]013);国家自然科学基金项目(41876204);广州市基础研究计划项目(202201011687);2023年广东省普通高校重点领域专项项目(2023ZDZX4085)。

摘　　要：针对MODIS不同波段之间信息冗余对叶绿素a(Chl)浓度建模的影响,该文在建模中引入主成分分析(PCA)方法,提出了一种PCA与BP神经网络相结合的Chl浓度反演模型(PCA-BPN)。通过主成分分析,从多个相关波段中提取出几个相互独立的关键主成分,然后将这些关键主成分作为BP神经网络的输入,通过网络自主学习构建Chl浓度反演模型。实验表明,前3个主成分包含了波段信息的99.5%,降低了神经网络的输入维度。与Aqua卫星上MODIS(MODISA)标准Chl反演模型OCI相比,PCA-BPN模型提高了反演精度,在全球海域Chl浓度反演中具有一定的应用潜力。Aiming at the correlation between MODIS different bands and the impact of information redundancy between different bands on Chl concentration modeling,the principal component analysis(PCA)method was introducec into the modeling and a Chl concentration inversion model combining PCA and BP network(PCA-BPN)was proposed in this paper.Through principal component analysis,several independent key principal components were extracted from multiple related bands,and then these key principal components were used as the input of BP network to construct the Chl concentration inversion model through network autonomous learning.Experimental results showed that the first three principal components contained 99.5%of the band information,which reduced the input dimension of the neural network.Compared with the MODISA standard Chl inversion model OCI,PCA-BPN model had higher inversion accuracy and certain application potential in the retrieval of Chl concentration in the global ocean water.

关 键 词：MODIS卫星 主成分分析 神经网络 叶绿素浓度 反演模型 

分 类 号：P237[天文地球—摄影测量与遥感]