高光谱成像技术的柑橘植株叶片含氮量预测模型

氮素是果树生长发育的一种大量必需元素，及时准确地监控果树的氮营养状况，对果树的合理施肥、增产、优化果实品质以及减缓过量施氮引起的水资源污染具有重要意义。利用高光谱成像技术结合多变量统计学方法，建立了柑橘植株叶片的含氮量预测模型。研究步骤为：高光谱扫描、提取平均光谱曲线、预处理原始光谱数据、采用连续投影法提取特征波段和建立含氮量预测模型。从SG平滑、SNV、MSC、1-Der等11种预处理方法中筛选出的较优预处理方法是SG平滑、Detrending和SG平滑-Detrending。对应这三种最优预处理方法，先采用连续投影法挑选出各自的特征波长，然后将各特征波段下的光谱反射率作为偏最小二乘、多元线性回归和反向传播人工神经网络模型的输入，各自建立三个预测模型。从以上获得的9个预测模型中，得出两个最优模型SG平滑-Detrending-SPA-BPNN(R_p：0.851 3，RMSEP：0.188 1)和Detrending-SPA-BPNN(R_p：0.8609，RMSEP：0.159 5)。结果表明，利用高光谱数据测定柑橘叶片含氮量具有可行性。这为实时、准确地监控柑橘植株生长过程中叶片含氮量的变化以及合理科学的氮肥施加提供了一定的理论基础。

Development of Prediction Models for Determining N Content in Citrus Leaves Based on Hyperspectral Imaging Technology

The present study presents prediction models for determining the N content in citrus leaves by using hyperspectral imaging technology combined with several chemometrics methods. The steps followed in this study are: hyperspectral image scanning, extracting average spectra curves, pretreatment of raw spectra data, extracting characteristic wavelengths with successive projection algorithm and developing prediction models for determining N content in citrus leaves. The authors obtained three optimal pretreatment methods through comparing eleven different pretreatment methods including Savitzky-Golay(SG)smoothing, standard normal variate(SNV), multiplicative scatter correction(MSC), first derivative(1-Der) and so on. These selected pretreatment methods are SG smoothing, detrending and SG smoothing-detrending. Based on these three pretreatment methods, the authros first extracted the characteristic wavelengths respectively with successive projection algorithm, and then used the spectral reflectance of the extracted characteristic wavelengths as input variables of partial least squares regression (PLS), multiple linear regression (MLR) and back propagation neural network (BPNN) modeling. Hence, the authors developed three prediction models with each pretreatment method, and obtained nine models in total. Among all the nine prediction models, the two models based on the methods of SG smoothing-detrending-SPA-BPNN (R_p：0.851 3，RMSEP：0.188 1)and detrending-SPA-BPNN (R_p：0.860 9，RMSEP：0.159 5)were found to have achieved the best prediction results. The final results show that using hyperspectra data to determine N content in citrus leaves is feasible. This would provide a theoretical basis for real-time and accurate monitoring of N content in citrus leaves as well as rational N fertilizer application during the plants growth.