基于多元统计分析的黑胫病烟草叶绿素含量高光谱估测

Hyperspectral Estimation of Chlorophyll Content of Tobacco Influenced by Phytophthora parasitica var. nicotianae Based on Multivariate Statistical Analysis

为快速有效地评估烟草病害等级,需要合理准确地估计烟草叶绿素含量。分别利用ASD Field Spec4地物光谱仪和HSY-051叶绿素仪测定黑胫病烟草旺长期叶片高光谱及其SPAD值,以叶片光谱原始反射率及其8种变换处理分别与叶绿素含量进行相关分析,然后利用多元统计分析方法(逐步回归和主成分回归)分别建立叶绿素含量估测模型。结果表明:1) 原始光谱的8种变形与叶绿素含量的相关性都高于原始光谱反射率;2) 各变换的一阶导数和二阶导数与叶绿素含量的相关性都明显高于其原始形式;3) 利用逐步回归方法建立的模型估测效果最好,其决定系数R2为0.8715。经验证,模型精度较好。本研究可为高光谱技术监测烟草叶片叶绿素含量提供一定的参考。

The content of chlorophyll is a very important assessment index in the growing situation of to-bacco. Thus, estimating the content of chlorophyll in tobacco accurately is effective for rating to-bacco disease. By use of the ASD Field Spec4 spectrometer to determine hyperspectral data of tobacco leaves infected by Phytophthora parasitica var. nicotianae in the vigorous growing period and HSY-051 chlorophyll meter to measure the SPAD value. Then, the correlation analysis of chlorophyll content with hyperspectral reflectance and its eight transforms was proceeded. On this basis, the chlorophyll content estimation models were established with the method of multivariate statistical analysis (stepwise regression and principal component regression). The results indicate that 1) the correlation between chlorophyll content and the eight transforms of original spectra was higher than that of the original spectral reflectance;2) the correlation between chlorophyll content and the first derivative or second derivative of each transform was significantly higher than that of original form;3) the best model was: Y = 22.415 + 39249.31X1 − 13943.06X2 − 24807.33X3 + 296102.44X4 − 309086.89X5 + 130909.69X6 − 7751.45X7, and this model with a correlation coefficient of 0.9335 and R2 of 0.8715 was based on stepwise regression. After verification, the precision of the model is good. Therefore, the research will provide a reference for the better application of hyperspectral technology in monitoring the chlorophyll content of tobacco leaves.