基于集成学习的砂姜黑土含水量高光谱反演研究

王志刚; 黄子琪; 贺成龙; 蔡太义; 冯玉庆; 陆宁静; 窦焕衡

文章摘要

基于集成学习的砂姜黑土含水量高光谱反演研究

Hyperspectral inversion study of Vertisol soil moisture content based on ensemble learning

Received:March 23, 2023 Revised:August 24, 2023

DOI：10.13254/j.jare.2023.0186

中文关键词: 土壤含水率，高光谱，砂姜黑土，堆叠集成，偏最小二乘回归，支持向量机回归

英文关键词: soil moisture content, hyperspectral, Vertisol, stacking ensemble, partial least squares regression, support vector machine regression

基金项目:河南省高校人文社会科学一般项目（2024ZZJH147）；国家自然科学基金项目（41671225）; 河南省重大科技专项（181100110400）

Author Name	Affiliation	E-mail
WANG Zhigang	School of Surveying and Land Information Engineering, Henan Polytechnic University, Jiaozuo 454000, China
HUANG Ziqi	College of Geographical Science, Harbin Normal University, Harbin 150025, China
HE Chenglong	School of Surveying and Land Information Engineering, Henan Polytechnic University, Jiaozuo 454000, China
CAI Taiyi	School of Surveying and Land Information Engineering, Henan Polytechnic University, Jiaozuo 454000, China	caity2008@ hpu.edu.cn
FENG Yuqing	School of Surveying and Land Information Engineering, Henan Polytechnic University, Jiaozuo 454000, China
LU Ningjing	School of Surveying and Land Information Engineering, Henan Polytechnic University, Jiaozuo 454000, China
DOU Huanheng	School of Surveying and Land Information Engineering, Henan Polytechnic University, Jiaozuo 454000, China

Hits: 857

Download times: 702

中文摘要:

为提高砂姜黑土土壤水分的估测精度，本研究以河南省西平县砂姜黑土为研究对象，通过配制不同含水率土壤样本并在室内进行高光谱测量，对土壤样本高光谱数据平滑（SR）、倒对数[LOG（1/R）]、一阶微分（FD）、多元散射校正（MSC）、去包络线（CR）光谱变换后，结合连续投影算法（SPA）识别最佳特征波段，采用偏最小二乘回归（PLSR）、支持向量机回归（SVR）的机器学习方法和堆叠（Stacking）集成学习方法分别构建土壤含水率反演模型。结果表明：经MSC变换后光谱中土壤含水率相关信息增强最多；SPA算法能对砂姜黑土含水率光谱数据进行降维和特征信息提取；经反射光谱MSC变换后由PLSR和SVR集成的Stacking集成模型决定系数最高（R²=0.963）、均方根误差最小（RMSE=1.7）。研究表明，Stacking集成学习模型有效提升了模型的精度和泛化能力，是砂姜黑土含水率最佳反演模型。

英文摘要:

To improve the accuracy of soil moisture estimation in Vertisols, this study took the Vertisol in Xiping County, Henan Province, China, as its research object and conducted hyperspectral measurement in the laboratory by configuring soil samples with different moisture contents after implementing smoothing（SR）, logarithm of the inverse[LOG（1/R）], first-order differentiation（FD）, multiple scattering correction（MSC）, and continuum removal（CR）spectral transformation processes on the soil sample hyperspectral data. The best feature bands were identified by combining the successive projection algorithm（SPA）with the machine learning methods of partial least squares regression（PLSR）and support vector machine regression（SVR）and stacking（Stacking）integrated learning methods were used to construct the soil water content inversion model. The results showed that the information related to soil water content was most enhanced in the MSC-transformed spectra. The SPA algorithm was able to downscale and extract feature information from the water content spectral data of the Vertisol. The Stacking integrated model, which integrated PLSR and SVR after MSC transformation based on the reflection spectra, had the highest coefficient of determination（R²=0.963）and the lowest root mean square error（^RMSE=1.7）. This study indicates that the Stacking integrated learning model is the best inversion model for Vertisol moisture content. It effectively improves the accuracy and generalization ability of the model.

HTML View Full Text View/Add Comment Download reader