基于光谱指数的苹果叶片水分含量估算模型研究

Hyperspectral Estimating Leaf Water Contents Based on Spectral Index in Apple

建立快速、无损的苹果叶片水分含量高光谱估算模型,为苹果树干旱预警提供理论依据。以2个不同生育期采集的苹果叶片为研究对象,研究了不同水分含量的苹果叶片高光谱特征,分析了苹果叶片水分含量与光谱指数之间的相关关系,建立了苹果叶片水分含量估算模型。结果表明,苹果叶片水分含量的敏感光谱波段主要集中于近红外和短波红外波段;利用6个光谱指数建立的单变量估算模型均达到了极显著水平(P<0.01),但以水分指数建立的估算模型y=29503x2-57746x+28317的拟合决定系数R2最大,为0.5401;经检验,拟合方程的RMSE为2.4,RE为5.8%,检验精度达到了94.2%。采用主成分回归分析方法,建立的苹果叶片水分含量估算模型y=-556.819+347.838x1-17.815x2-27.864x3+299.492x4+25.647x5+9.835x6的拟合决定系数R2为0.6371,经检验,拟合方程的RMSE为1.26,RE为1.8%,检验精度达到了98.2%。表明以主成分回归分析建立的苹果叶片水分含量估算模型具有较好的敏感性和稳定性。

The objective of the paper is to establish a hyperspectral estimating model of rapid and nondestructive for the apple leaf water contents and to provide the theoretical basis of drought warning for the fruit tree. The apple leaves were collected as the research object on two different growthes and studied the different water contents of apple leaves of hyperspectral features. The relationship was analyzed between the different leaves water contents and spectral parameters. Estimating models of apple leaf water contents were set up by means of constructing six common spectral indexes which were WI, WBI, PWI, GVWI, MSI and NDWI. The results indicated that sensitive bands of the apple leaf hyperspectral were focused mainly on the near infrared and short wave infrared light areas. All single variable estimating models reached a significant level （P〈0.01） by six vegetation moisture indexes. But the moisture index established estimation model was y=29503x2-57746x ＋28317, its fitting determination coefficient/T=0.5401, was the largest. The model tested,its RMSE was 2.4 and RE was 5.8%. The test accuracy of the model reached 94.2%. Using principal component analysis method, the model was y=-556.819＋347.838x1-17.815x2-27.864x3＋299.492x4＋25.647x5＋ 9.835x6, its fitting determination coefficient R2=0.6371. The model tested, its RMSE was 1.26 and RE was 1.8%. The test accuracy of model reached 98.2%. It showed that the estimating model of apple leaves water contents had good sensitivity and stability using principal component regression analysis method. Key words： spectral index; apple leaf; water content; estimating model