引用本文:沈兰芝,高懋芳※,闫敬文,姚艳敏.基于SVR 和PLSR 的土壤有机质高光谱估测模型研究[J].中国农业信息,2019,31(1):58-71
【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 104次   下载 48 本文二维码信息
码上扫一扫!
分享到: 微信 更多
基于SVR 和PLSR 的土壤有机质高光谱估测模型研究
沈兰芝1, 高懋芳※2, 闫敬文1, 姚艳敏2
1.汕头大学工学院,广东汕头515063;2.中国农业科学院农业资源与农业区划研究所/ 农业农村部农业遥感重点实验室,北京100081
摘要:
【目的】探讨高光谱遥感数据不同预处理及不同估测算法下土壤有机质估测模型的优 劣,为提高土壤有机质估测精度奠定基础。【方法】使用高光谱仪在室内条件下对土壤样品 进行光谱测量,对光谱数据进行4 种去噪处理(无去噪处理、Savitzky-Golay(S-G)平滑滤 波去噪、小波包去噪以及S-G 平滑与小波包结合去噪),然后对去噪后的光谱数据进行8 种 数据变换(原始光谱数据R、倒数1/R、对数log(R)、倒数对数log(1/R)、一阶导数R′、 倒数一阶导数(1/R)′、对数一阶导数(log(R))′、倒数对数一阶导数(log(1/R))′), 接着对变化后的光谱数据进行3 种降维处理(无降维处理、敏感波段降维和主成分分析降 维),最后运用支持向量回归法和偏最小二乘回归法分别建立SOM 含量估测模型。【结果】 研究中所涉及的各种数据预处理和估测算法中,小波包去噪、PCA 降维、反射率倒数一阶导 数(1/R)′ 光谱数据变换处理条件下,使用PLSR 方法的估测模型精度最高、模型最稳定, 可以较精确地估测吉林省伊通县SOM 含量。【结论】合适的数据预处理,尤其是小波包去噪 和PCA 降维相结合,可有效改善光谱数据质量,提高SOM 含量估测模型精度及稳定性。
关键词:  土壤有机质;支持向量回归;偏最小二乘回归;小波包去噪;PCA 降维;高光谱
DOI:10.12105/j.issn.1672-0423.20190106
分类号:
基金项目:国家自然科学基金项目“耦合遥感与作物生长模型的农业干旱预警研究”(41871282);中国地质调查工作 项目(DD20160068);高分辨率对地观测系统国家科技重大专项(09-Y30B03-9001-13/15)
Estimation model of soil organic matterbased on SVR and PLSR
Shen Lanzhi1, Gao Maofang※2, Yan Jingwen1, Yao Yanmin2
1.College of Technology,Shantou University, Guangdong Shantou 515063, China;2.Institute of Agricultural Resources and Regional Planning,Chinese Academy of Agricultural Sciences/Key Laboratory of Agricultural Remote Sensing, Ministry of Agriculture,Beijing 100081, China
Abstract:
[Purpose]The advantages and disadvantages of Soil Organic Matter( SOM)estimation models under different pretreatment and different estimation algorithms for hyperspectral remote sensing data are discussed,which lays a foundation for improving soil organic matter estimation accuracy.[Method]In our work,the spectra of soil samples were measured in laboratory using a high spectral spectrometer. Four kinds of denoising methods (Non-denoising,Savitzky-Golay( S-G) smoothing filtering,wavelet packet denoising and S-G smoothing combined with wavelet packet denoising)were used to process the spectral data. Eight kinds of spectral data transformations( R,1/R,log(R),log(1/R),R′,(1/R)′,(log(R))′ and( log(1/R))′)are performed on the denoised spectral data. And three kinds of dimensionality reduction processing( Non-dimensionality reduction,sensitive band dimensionality reduction and Principal Component Analysis( PCA) dimensionality reduction) are carried out on the changed spectral data. Finally,the SOM content estimation model was established by Support Vector Regression( SVR) and Partial Least Square Regression( PLSR).[Result]Among the various data preprocessing and estimation algorithms involved in this work, wavelet packet denoising,PCA dimensionality reduction,and reflectance first derivative( 1/R)′ spectral data transformation has the highest accuracy and stability under the model that established by PLSR,which can accurately estimate the SOM content of Yitong County,Jilin Province.[Conclusion]Appropriate data preprocessing,especially the combination of wavelet packet denoising and PCA dimensionality reduction,can effectively improve the quality of spectral data and improve the accuracy and stability of SOM content estimation model.
Key words:  SOM;SVR;PLSR;wavelet packet denoising;PCA dimension reduction;hyperspectral