Accurate monitoring of crop moisture content is very important for irrigation scheduling and yield increase. This study aims to construct an optimal estimation model of winter wheat leaf moisture content (LMC) through spectral data processing and feature band selection. LMC and spectral reflectance were measured in 2017-2018 to construct models using simple linear regression (SLR), principal components regression (PCR), and partial least square regression (PLSR); feature bands for modelling were selected through correlation analysis and the effects of feature band number on estimation accuracy were compared. The results showed that data transformation significantly enhanced the correlation between spectral features and LMC. However, the band position corresponding to the maximum correlation coefficient for each transformation was not fixed. The accuracy of PLSR models were significantly higher than that of PCR and SLR models. The comparison of relative percent deviation (RPD) values indicated that the RPD values increased rapidly and then tended to be stable with the increase of feature band number. The R′′ -PLSR model constructed with 28 feature bands (R²_c = 0.8517; RPD > 2.0) estimated the LMC more accurately than other models. This study provides a good method for non-destructive monitoring of crop moisture content.

准确监测农作物水分含量对于灌溉计划和增产非常重要。本研究旨在通过光谱数据处理和特征带选择构建冬小麦叶片含水量的最佳估计模型。在2017-2018年测量LMC和光谱反射率，以使用简单线性回归（SLR），主成分回归（PCR）和偏最小二乘回归（PLSR）构建模型; 通过相关性分析，选择用于建模的特征带，并比较特征带数目对估计精度的影响。结果表明，数据转换显着增强了光谱特征与LMC之间的相关性。然而，对应于每个变换的最大相关系数的带位置不是固定的。PLSR模型的准确性显着高于PCR和SLR模型。相对百分比偏差的比较（RPD）值表明RPD值迅速增加，然后随着特征带数的增加趋于稳定。与其他模型相比，由28个特征带（R ² _c = 0.8517；RPD > 2.0）构建的R ''-PLSR模型估计的LMC更准确。这项研究为作物水分含量的无损监测提供了很好的方法。