谷子叶绿素含量高光谱特征分析及其反演模型构建

彭晓伟; 张爱军; 杨晓楠; 王楠; 赵丽

彭晓伟,张爱军,杨晓楠,王楠,赵丽.谷子叶绿素含量高光谱特征分析及其反演模型构建[J].干旱地区农业研究,2022,40(2):69~77

谷子叶绿素含量高光谱特征分析及其反演模型构建

Hyperspectral characteristics and remote sensing inversion model of chlorophyll content of millet

DOI：10.7606/j.issn.1000-7601.2022.02.09

中文关键词: 谷子叶绿素含量高光谱特征波段反演模型

英文关键词:millet chlorophyll content hyperspectral characteristic band inversion model

基金项目:河北省重点研发计划项目（19226421D）

作者	单位
彭晓伟	河北农业大学国家北方山区农业工程技术研究中心，河北保定 071000
张爱军	河北农业大学国家北方山区农业工程技术研究中心，河北保定 071000；河北省山区研究所，河北保定 071000
杨晓楠	河北农业大学国家北方山区农业工程技术研究中心，河北保定 071000
王楠	河北农业大学机电工程学院，河北保定 071000
赵丽	河北农业大学农学院，河北保定 071000

摘要点击次数: 556

全文下载次数: 417

中文摘要:

基于高光谱数据综合分析不同施肥条件下谷子各生长期冠层叶绿素含量的高光谱特征，在分析各光谱特征参数与叶绿素相关性的基础上，基于偏最小二乘法和人工神经网络构建叶绿素含量的遥感反演模型。结果表明：NDVI（归一化植被指数）、GNDVI（绿色归一化植被指数）、PSNDa（特殊色素归一化指数a）、PSSRc（特征色素简单比值指数c）、RENDVI（红边归一化植被指数）及Dy（黄边幅值）与不同生育期的SPAD值均呈极显著相关关系（P<0.05）。基于上述光谱指数为自变量建立的最佳一元回归模型R²（决定系数）在0.4~0.6之间，基于偏最小二乘法的回归模型R²在0.55~0.71之间，RMSECV（交叉验证均方根）在1.34~2.23之间，Q²_cum（主成分累积模型预测能力）在0.54~0.83之间，对自变量的解释能力在63.1%~95.8%之间，说明上述光谱参数对叶片叶绿素的解释程度较好。利用BP神经网络估测叶绿素含量可达到最优精度，建模集的R²达到0.70以上，RMSE（均方根误差）在1.18~2.48之间。综上所述，利用BP神经网络建模效果最优。

英文摘要:

This study used the comprehensive analysis of the hyperspectral data of the hyperspectral characteristics of the chlorophyll content in the millet canopy under different fertilization conditions to examine the correlation between the spectral characteristics and the chlorophyll. The remote sensing inversion model of chlorophyll content was constructed based on the partial least squares method and artificial neural network. The results showed that: through correlation analysis, NDVI, GNDVI, PSNDa, PSSRc, RENDVI, and Dy all had extremely significant correlations with SPAD in different growth stages. The coefficient of determination R² of the best unary regression model established based on the above spectral index as the independent variable was between 0.4 and 0.6, and the coefficient of determination R² of the regression model based on the partial least squares method was between 0.55 and 0.71. The cross\|validated root mean square RMSECV fell between 1.34 and 2.23, and the predictive ability of the principal component accumulation model Q²_cum was between 0.54 and 0.83.The explanatory ability of the independent variable was between 63.1% and 95.8%, indicating that the above\|mentioned spectral parameters explained the leaf chlorophyll better. The BP neural network estimated the chlorophyll content to achieve the best accuracy, and the determination coefficient R² of the modeling set was above 0.70.The RMSE was between 1.18 and 2.48. In summary, the modeling effect using BP neural network was the best.

查看全文查看/发表评论下载PDF阅读器