基于分数阶微分的葡萄叶片SPAD值高光谱遥感反演研究  

Hyperspectral remote sensing inversion of SPAD values in grape leaves based on fractional differentiation

作　　者：郭松 舒田[1] 刘春艳 冯恩英 王文静 蒋丹垚 GUO Song;SHU Tian;LIU Chun-yan;FENG En-ying;WANG Wen-jing;JIANG Dan-yao(Guizhou Agricultural Science and Technology Information Institute,Guiyang 550006,China;College of Natural Resources and Envi-ronment,Northwest A&F University,Yangling,Shaanxi 712100,China)

机构地区：[1]贵州省农业科技信息研究所,贵阳550006 [2]西北农林科技大学资源环境学院,陕西杨凌712100

出　　处：《西南农业学报》2024年第2期446-456,共11页Southwest China Journal of Agricultural Sciences

基　　金：贵州省科学技术基金项目(黔科合基础⁃ZK〔2021〕一般130号);科研机构创新能力建设专项(黔科合服企〔2021〕15号)。

摘　　要：【目的】探明高光谱遥感技术反演葡萄叶片叶绿素含量的可能性,构建葡萄叶片叶绿素含量反演模型,为快速且无损估测葡萄长势提供技术参考。【方法】以西南山区成熟期葡萄叶片为研究对象,同步获取冠层叶片高光谱数据和SPAD值,研究不同分数阶(0.0~1.4阶,步长0.2阶)微分光谱反演葡萄叶片SPAD值的能力,构建多个基于特征波段和光谱指数的单因素模型及基于连续投影算法的多因素模型。【结果】不同SPAD值葡萄叶片原始光谱曲线整体一致,在可见光区域反射率较低而在近红外区域反射率高;可见光、近红外区域反射率与SPAD值分别呈反比和正比;随着分数阶上升,特征波段由近红外向红边靠近,光谱指数由近红外与蓝光组合变更为近红外与绿光组合,单因素模型建模变量相关性呈先升后降趋势,在0.6阶达峰值;除0.6与0.8阶外,其余分数阶微分光谱单因素模型建模变量均为DSI;多因素模型优于单因素模型,机器学习算法可提升传统回归模型精度,所有模型以0.6阶下SPA⁃GA⁃XGBoost回归模型精度最优,其建模与验证R^(2)分别为0.79和0.75,相应均方根误差(nRMSE)分别为15.54%和14.45%。【结论】分数阶微分变换在葡萄叶片SPAD值反演方面具有较大潜力,特定分数阶下,光谱指数优于特征波段,GA⁃XGBoost算法能产生较好的建模效果。【Objective】The possibility of inverting chlorophyll content of grape leaves by hyperspectral remote sensing technology was ex⁃plored,and the inversion models of chlorophyll content of grape leaves were established,which provided the technical references for rapid and non⁃destructive estimation of the growth potential of grape.【Method】The grape leaves at maturity stage in the southwestern mountainous areas were used as the research objects,and the hyperspectral data and SPAD values of canopy leaves were obtained simultaneously,and then the ability of different fractional differention(0.0-1.4 order,step size 0.2)spectrum inverting SPAD values of grape leaves was studied,and finally the multiple single⁃factor models based on feature bands and spectral index and multi⁃factor models based on successive projections algorithm were constructed.【Result】The original spectral curves of grape leaves with different SPAD values were consistent as a whole,and the reflectance of grape leaves was low in the visible light region and high in the near⁃infrared region.The reflectance of visible light and near⁃infrared region was inverse proportion and proportion to SPAD value respectively.With the increase of the order,the feature bands moved from near infrared to red edge,the spectral index changed from the combination of near infrared and blue light to the combina⁃tion of near infrared and green light,and the correlation of the modeling variables of the single⁃factor models first increased and then decreased,reaching a peak value at order 0.6.Except for order 0.6 and 0.8,the modeling variables of single⁃factor models of other fractional differentiation were DSI.The multi⁃factor models were superior to the single⁃factor models,and the machine learning algorithm could improve the accuracy of the traditional regression model.Among all the models,the SPA⁃GA⁃XGBoost regression model with the best accuracy was the 0.6⁃order model,whose modeling and verification R^(2)were 0.79 and 0.75,respectively,

关 键 词：葡萄叶片 SPAD 分数阶微分 光谱指数 GA⁃XGBoost回归 

分 类 号：S127[农业科学—农业基础科学]