胡杨叶片水分含量是评价胡杨健康状况的重要指标。 光谱检测法是一种常用的手段, 但在近红外光谱的测量过程中, 在一定程度上必然受到仪器噪声、 摆放形态差异和环境的干扰, 为避免噪声、 散射对近红外光谱的影响, 减少数据维数, 采用多元散射校正（MSC）算法对原始光谱数据进行预处理, 去除散射和基线漂移的影响, 增加了光谱数据的信噪比, 使有效光谱信息较为明显, 谱带特征得到加强, 有利于特征波长的选择。 为降低模型的复杂度, 防止过拟合现象, 减小共线性影响, 利用连续投影算法（SPA）进行特征变量选择, 并通过多元线性回归模型, 分析各个波长模拟的残差平方, 评价各个波长的贡献, 剔除贡献较小的波长, 最终获得用于建模的特征波长, 改善建模条件。 最后使用偏最小二乘回归算法建立胡杨叶片水分含量检测模型。 实验表明, 直接使用原始光谱, 利用SPA算法筛选变量个数为6个, 模型预测精度为90.144%, 相关系数r=0.674 24, RMSE=0.021 434, MSC处理后, 利用SPA算法选定最终变量数为5个, 预测精度为97.734%, 相关系数r=0.781 63, RMSE=0.016 776。 MSC和SPA算法有效的消除了散射噪声、 减小了共线性干扰, 模型的预测精度和相关性增加, 误差减小, 可用于胡杨叶片水分的快速无损检测, 而且对其他作物叶片水分检测也具有一定的借鉴意义。

The moisture content of leaves is an important index to evaluate the health condition of the Euphrates poplar, and spectrum detection method is an effective method. But in the process of near infrared spectrum measurement, spectral data will be affected by instrument noise, morphological differences and environment interference. The paper proposed how to establish spectrum detection method of water content of populus euphratica leaf. In the first place, the influence of scattering, noise and baseline drift of spectral data are reduced by using multiple scattering correction (MSC）, and increase signal to noise ratio (SNR) of spectrum data andstrengthen band features. The effective spectral information is relatively clear, so the choice of characteristic wavelength becomes easier. Then, in order to reduce the complexity of the model, to prevent overfitting and to reduce the influence of collinearity, the successive projections algorithm (SPA) is used to select the feature variables. And a multiple linear regression model is used to analysis and compare the simulated residual squared of different models, evaluate the contribution of each wavelength and eliminate the wavelengths of small contribution value. Then we obtained the optimal characteristic wavelength to improve the conditions of modeling. Finally, the partial least squares regression method is used to establish the test model. The experimental results show that the successive projections algorithm screens six effective variables on the basis of using the original spectrum data, the prediction accuracy is 90.144%, the correlation coefficient (r) is 0.674 24, the root mean square error is 0.021 434, and the successive projections algorithm screens five effective variables after using multiple scattering correction algorithm to optimize the original spectrum data, the prediction accuracy is 97.734%, the correlation coefficient (r) is 0.781 63, the root mean square error is 0.016 776. Sothe multiple scattering correction algorithm and successive projections algorithm has been successfully applied to eliminate the scattering noise, reduce the total linear interference, simplify the complexity of the model, then increase the accuracy and correlation coefficient, reduce the error. This method can be used for fast nondestructive testing of water content of the Euphrates poplar leaf, besides it also has some reference significance for moisture detection of other crops leaf.