投稿中心

审稿中心

编辑中心

期刊出版

网站地图

友情链接

引用本文:刘忠华,李云梅,吕恒,徐祎凡,徐昕,黄家柱,檀静,郭宇龙.基于偏最小二乘法的巢湖悬浮物浓度反演.湖泊科学,2011,23(3):357-365. DOI:10.18307/2011.0307
LIU Zhonghua,LI Yunmei,LV Heng,XU Yifan,XU Xin,HUANG Jiazhu,TAN Jing,GUO Yulong.Inversion of suspended matter concentration in Lake Chaohu based on Partial Leastsquares Regression. J. Lake Sci.2011,23(3):357-365. DOI:10.18307/2011.0307
【打印本页】   【HTML】   【下载PDF全文】   查看/发表评论  【EndNote】   【RefMan】   【BibTex】
←前一篇|后一篇→ 过刊浏览    高级检索
本文已被:浏览 3535次   下载 1670 本文二维码信息
码上扫一扫!
分享到: 微信 更多
基于偏最小二乘法的巢湖悬浮物浓度反演
刘忠华, 李云梅, 吕恒, 徐祎凡, 徐昕, 黄家柱, 檀静, 郭宇龙
南京师范大学虚拟地理环境教育部重点实验室, 南京 210046
摘要:
悬浮物浓度是水质评价的重要参数.对2009年6月巢湖实测的高光谱数据进行小波变换以去除数据冗余,减少建模时间.考虑到不同的小波基函数和分解尺度对数据压缩的影响,采用3个指标作为评价标准,最终选择小波基函数为Db4,分解尺度为4的小波变换,经小波变换后,原来的451个波段的光谱数据压缩为34个特征变量.利用小波变换后的光谱数据建立了悬浮物浓度反演的偏最小二乘法(PLS)模型,其中20个样本用于建模,9个样本用于验证,结果表明:当主成分个数为3时,PLS模型具有最佳的效果,此时模型的R2为0.93,R2(pred)为0.89,PRESS为3.29,3个主成分累积解释了98.60%的自变量信息和92.37%的因变量信息.此外,PLS模型能够充分利用高光谱数据信息,具有较高的精度和稳定性(R2=0.93,RMSE=4.77mg/L,MAPE==9.02%).通过与单波段模型、光谱一阶微分模型及波段比值模型的对比分析得出,PLS模型无论是从建模样本精度还是验证样本的误差方面均高于传统的经验模型,适合于利用高光谱数据进行悬浮物浓度的反演.
关键词:  小波变换  偏最小二乘法  高光谱数据  悬浮物  巢湖
DOI:10.18307/2011.0307
分类号:
基金项目:高分辨率对地观测系统国家科技重大专项项目(E0203/1112/JC01);国家自然科学基金项目(40971215);江苏省2008年度普通高校研究生科研创新计划(CX09B-301Z);南京师范大学优秀博士论文培养计划项目(12432116011036)联合资助
Inversion of suspended matter concentration in Lake Chaohu based on Partial Leastsquares Regression
LIU Zhonghua, LI Yunmei, LV Heng, XU Yifan, XU Xin, HUANG Jiazhu, TAN Jing, GUO Yulong
Key Laboratory of Virtual Geographic Environment of Education Ministry, Nanjing Normal University, Nanjing 210046, P. R. China
Abstract:
Suspended mailer concentration is an important parameter of water quality evaluation.Hyperspeetral data measured in lake Chaohu in June, 2009 were processed by wavelet transform in order to remove data redundancy and reduce modeling time. Three evaluation indexes were selected considering the effect of different wavelet functions and decomposed scales on the data compression.and the wavelet function Db4 and decomposed scale 4 were determined finally.The orional hyperspeetml data of 451 bands were compressed to 34 feature variables by the wavelet transform.Then 20 samples were used to construct Partial Least-squares Regression(PLS) inversion model of suspended matter concentration, and other 9 samples were used for model verification. The results show that PLS model is suited when the number of principal components is 3 and its R2 is 0.93, R2(pred) is 0.89 and PRESS is 3.29.These three principal components explain 98.60% of independent variables information and 92.37% of dependent variables information.PLS model with R2 of 0.93, RAISE of 4.77mg/L, and MAPE of 9.02% can make full use of the information of hyperspeetral data, and hence have higher accuracy and stability.In addition, single band model, spectral one-order differential model and band ratio model were used to compare with PLS model.The results show that PLS model is better than traditional empirical models no matter on the accuracy of modeling samples or the error of validation samples, indicating that it is suitable for the inversion of suspended matter concentration by using hyperspectral data.
Key words:  Wavelet transform  Partial Least-squares Regression  hyperspectral data  suspended matter  Lake Chaohu
分享按钮