基于光谱变换与SPA-SVR的玉米SPAD值高光谱估测  

Hyperspectral estimation of maize SPAD value based on spectrum transformation and SPA-SVR

作　　者：郭松 常庆瑞[1] 崔小涛 张佑铭 陈倩 蒋丹垚 落莉莉 GUO Song;CHANG Qingrui;CUI Xiaotao;ZHANG Youming;CHEN Qian;JIANG Danyao;LUO Lili(School of Natural Resources and Environment,Northwest A&F University,Yangling Shaanxi 712100,China)

机构地区：[1]西北农林科技大学资源环境学院,陕西杨凌712100

出　　处：《东北农业大学学报》2021年第8期79-88,共10页Journal of Northeast Agricultural University

基　　金：国家自然科学基金项目(41701398)。

摘　　要：叶绿素含量快速、无损监测是评估玉米生长状态有效方式之一。以抽雄期玉米为研究对象,研究原始光谱、普通一阶导数光谱、间隙一阶导数光谱、开平方根光谱以及连续统去除光谱的特征波段以及5个传统植被指数与玉米叶绿素含量之间关系。对比分析不同模型(单因素回归模型、结合连续投影与多元线性回归、支持向量回归模型)对抽雄期玉米叶绿素含量预测能力。结果表明,光谱变换可增强特征波段与SPAD值相关性,同时还增加敏感波段数量、提升建模精度;连续投影算法对特征降维效果明显,各类多元模型最优光谱参数为5~9个;各类型光谱下均为多因素模型精度优于单因素模型。其中,基于普通一阶导数光谱的支持向量回归模型为最优模型,其建模R^(2)与验证R^(2)分别达到0.92与0.90。光谱变换在反演玉米叶绿素方面有较大潜力,连续投影与支持向量回归结合可产生较好建模效果。Rapid and non-destructive estimation of chlorophyll is an effective way for evaluating the growing state of maize.This study took maize at tasseling stage as the research object,investigating the relationships between the measure value of chlorophyll of maize and original canopy spectrums and transformed spectrums including the ordinary first derivative of original spectrum,gap first derivative spectrum,square root of spectrum,the continuum removal spectrum and five traditional vegetation indices,comparing and analyzing the chlorophyll forecasting abilities of different models,including single factor regression model,multi-factors linear regression model combined with successive projections,support vector regression model.The results demonstrated that the transformed spectrums not only enhanced the correlations between themselves and SPAD value of maize,but also increased the number of sensitive feature bands and was helpful for improving accuracy of forecasting models.The successive projections algorithm performed very well on dimensionality reduction,and the optimal number of spectrum features of each multi-factors model used ranged from five to nine.The accuracies of the multi-factor models were prior to those of the single-factor models no matter which features were used to construct these models.The support vector regression model based on the ordinary first derivative spectrum was the best model,with modeling R^(2) and validation R^(2) were 0.92 and 0.90,respectively.The research proved that transformed spectrums were of great potential in the inversion of maize chlorophyll,and integrating successive projections and support vector regression was respected as a feasible and valuable approach to forecast SPAD value of maize.

关 键 词：玉米 SPAD值 光谱变换 连续投影算法 支持向量回归 

分 类 号：S127[农业科学—农业基础科学] S513