基于冠层光谱的早籼稻籽粒粗蛋白含量估测

利用2019年长江中下游早籼稻早中熟组种质资源为材料,分析主要生育期冠层光谱反射率与籽粒粗蛋白含量的关系,筛选出可用于早籼稻籽粒粗蛋白含量预测的敏感生育期和敏感波长,建立了基于敏感波长和光谱参数的籽粒粗蛋白的一元线性、多元线性、指数和多项式预测模型,用决定系数（R ²）、均方根误差（RMSE）和相对误差（RE）对模型精度进行评价,以期找到估测早籼稻籽粒粗蛋白含量的最适模型。研究发现,在孕穗期514、580、638和695nm波长处冠层一阶微分光谱反射率与籽粒粗蛋白含量相关性达到极显著水平;在基于敏感波长的估测模型中,四元线性模型估测效果最佳,其建模集R ²、RMSE和RE分别为0.566、0.342%和2.874%,验证集R ²、RMSE和RE分别为0.518、0.154%和1.303%;在基于光谱参数构建的估测模型中,DSI(R′₅₁₄和R′₆₃₈)为自变量构建的多项式模型估测效果较优,其建模集R ²、RMSE和RE分别为0.638、0.312%和2.639%,验证集R ²、RMSE和RE分别为0.581、0.230%和2.307%。

Estimation of Crude Protein Content in Grain of Early Indica Rice Based on Canopy Spectrum

In order to find the most suitable model of estimating the crude protein content of grains, we took the germplasm resources of early and middle maturity groups of early indica rice varieties in the middle and lower regimes of the Yangtze River in 2019 as materials, analyzed the relationships between the canopy spectra reflectance at the main growth stages and the crude protein content of grains, and also screened out the optimal period and sensitive wave length that could be used to predict the crude protein content in the early indica rice varieties. The univariate linear, multivariate linear, exponential, and polynomial prediction models for the crude protein content of grains based on sensitive wave lengths and spectra parameters were established, and the determination coefficient (R ²), root mean square error (RMSE) and relative error (RE) were used to evaluate the accuracy of the models, for finding the best model to estimate the crude protein content of early indica rice. The results showed that the correlation between the first derivative spectral reflectance of the canopy and the crude protein content of the grain at 514, 580, 638 and 695nm reached an extremely significant level at booting stage. Among the estimation models based on sensitive wave length, the quaternary linear model had the best estimation effects, with 0.566, 0.342% and 2.874% for the calibration set R ², RMSE and RE, and 0.518, 0.154% and 1.303% for the validation set R ², RMSE and RE, respectively. In the estimation model that constructed based on the spectra parameters, the polynomial model which constructed by DSI (R′₅₁₄, R′₆₃₈) as the independent variable had a better estimation effects, and its calibration set R ², RMSE and RE were 0.638, 0.312% and 2.639%, respectively, the validation set R ², RMSE and RE were 0.581, 0.230% and 2.307%, respectively.