RSEI应使用主成分分析或核主成分分析?  

Should RSEI Use PCA or kPCA?

作　　者：徐涵秋[1,2] 李春强 林梦婧 XU Hanqiu;LI Chunqiang;LIN Mengjing(College of Environment and Safety Engineering,Institute of Remote Sensing Information Engineering,Fuzhou University,Fuzhou 350116,China;Fujian Provincial Key Laboratory of Remote Sensing of Soil Erosion,Fuzhou University,Fuzhou 350116,China;Institute of Urban Environment,Chinese Academy of Sciences,Xiamen 361021,China)

机构地区：[1]福州大学环境与安全工程学院,福州大学遥感信息工程研究所,福建福州350116 [2]福州大学福建省水土流失遥感监测评价与灾害防治重点实验室,福建福州350116 [3]中国科学院城市环境研究所,福建厦门361021

出　　处：《武汉大学学报（信息科学版）》2023年第4期506-513,共8页Geomatics and Information Science of Wuhan University

基　　金：国家自然科学基金(31971639)。

摘　　要：nRSEI(nonlinear remote sensing ecological index)是新近提出的遥感生态指数,它采用核主成分分析(kernel principal component analysis,kPCA)来集成模型的各个分指标。其主要根据是认为原RSEI采用的湿度、绿度、干度、热度这4个指标在北京研究区的相关关系总体为弱相关,因此需要采用专门处理非线性关系的kPCA来集成这4个指标。为此探讨了北京地区这4个指标的相关关系类型,并对新指数验证方法的有效性进行了深入分析。结果表明,北京地区这4个指标总体呈显著的强线性相关关系,因此并不适合采用kPCA集成;新指数的精度验证方法也存在明显的缺陷,不能证明新指数的有效性。同时还就遥感建模的可行性、模型的普适性、指标尺度的一致性,以及模型精度的验证方法、标准参考影像的选取和验证所需的样本量等遥感研究论文中常见的基础问题进行了讨论。Objectives The nonlinear remote sensing ecological index(nRSEI)is a recently proposed ecological index,which used the kernel principal component analysis(kPCA)algorithm to integrate four indicators of the existing remote sensing ecological index(RSEI)rather than using the traditional principal component analysis(PCA)technique.The main reason for using kPCA was that in the Beijing area the correlations between wetness,greenness,dryness,and heat that are four indicators used in RSEI were generally weak,so the kPCA that is specially developed to deal with variables with nonlinear relationship was needed to handle these four non-linear indicators.This paper aims to examine the correlation strength of these four indicators in the Beijing area to see whether their relationship is strong or weak and analyzes the effectiveness of the accuracy assessment method that was used for the validation of the new index.Methods Through examining correlation coefficients and scatter diagrams,the correlation between the four indicators is investigated to find out whether the relationship between the indicators is linear or non-linear.Also,the effectiveness of the validation for the new index is analyzed.Results The results show that the four indicators are strongly linearly correlated with each other,therefore,the kPCA was not suitable for the intergradation of the four indicators.The methods used for the accuracy assessment of the new index also have obvious defects and thus failed to effectively validate the accuracy of the new index.In addition,some important issues related to prepare remote sensing papers are also discussed.These include the feasibility of remote sensing modeling,the robustness of the model,the scale consistency of the sub-indicators in the model,the validation method for the model,the selection of reference images for the validation,and the required sample size for accuracy validation.Conclusions The nRSEI is not a suitable index for the assessment of regional ecological status as it mistakenly employed kPCA to

关 键 词：RSEI 主成分分析 核主成分分析 相关性强度 精度验证 

分 类 号：TP75[自动化与计算机技术—检测技术与自动化装置] X87[自动化与计算机技术—控制科学与工程]