Prediction of grain protein content using ear layer spectral reflectance data

صينى. 本试验对近地面高光谱仪监测小麦的不同测定方法进行了探讨，并比较了各方法对籽粒蛋白质含量（GPC）的预测能力。结果表明，穗全氮含量（ETNC）与穗层光谱反射率（Rel）的相关系数普遍高于与冠层光谱反射率（Rc）的相关系数。同时，基于小麦光谱反射率、穗全氮含量、籽粒蛋白质含量三者之间的相关性，选择了包括植被指数在内的20个穗层光谱特征参量，与ETNC进行相关分析，建立了最佳光谱特征参量预测ETNC以及ETNC预测籽粒蛋白质含量（GPC）的统计相关模型。通过2个模型的链接，建立了利用比值植被指数RVI[890，670]预测GPC的回归模型，可以较好地预测小麦籽粒蛋白质含量。在相同条件下，相对于以往基于冠层光谱的方法，基于穗层光谱的RVI[890，670]对CPC的预测表现出较大的优势，决定系数R. 由0.662提高到0.865，总均方根差RMSE由0.851降低到0.734。本研究为实现田间条件下小麦氮素及相关品质指标的便携式监测仪的开发研制提供了初步的依据。

إنجليزي. Remote sensing has been successfully applied in assessing crop biophysical and biochemical variables, such as leaf area index, chlorophyll, nitrogen and so on. But the grain quality non-destructive predicting is a new project focused on remote sensing application in agriculture. Some researchers have investigated quality prediction through estimating leaf nitrogen content with canopy reflected spectral characteristic parameters. But the canopy spectrum achieved by traditional method is a mixed spectrum affected by plant, soil and other factors, so from that it is difficult to extract the interested information. In this research, an improved method was used to determine grain protein content （GPC）. The results showed that the ear layer spectral reflectance （Rel） measured by the improved method was closer to single ear spectral reflectance （Re） than canopy spectral reflectance （Rc） by traditional method （Fig.2）. Based on the highly relativity of Rel,ear total nitrogen content （ETNC） and GPC, 20 spectral characteristics parameters were selected in this study, and the correlations between ETNC and them were analyzed respectively. It indicated that ear layer spectral parameters had stronger correlation with ETNC than canopy ones（Table 2）. The new model was established to predict GPC using ratio vegetation index （RVI [ 890,670 ] ）. The coefficient of determination increased from 0. 662 from Rel, to 0. 865 from Rc, and the Root mean square error（RSME） decreased from 0.851 to 0.734 （ Fig. 4）, which indicated that the prediction model using RVI [ 890,670 ] calculated from Rel is more reliable and practicable than that from canopy spectral reflectance. The results provide the theoretical foundation and technique support for measuring wheat nitrogeo status or quality parameters by aviation and spaceflight remote sensing, expecially expXoiting the portable testing meter for nitrogen.