水稻叶片氮素及籽粒蛋白质含量的高光谱估测模型

ESTIMATING NITROGEN OF RICE LEAF AND PROTEIN OF RICE SEED BASED ON HYPERSPECTRAL DATA

研究水稻叶片氮素和籽粒蛋白质含量的高光谱快速、无损监测方法,对于水稻营养诊断、籽粒品质监测及氮肥高效利用具有重要意义。本文通过水稻盆栽试验,测定水稻叶片氮素、籽粒蛋白质含量和冠层光谱,采用不同的光谱建模方法来提高氮素、籽粒蛋白质含量的估测精度。先用主成分分析(PCA)方法进行特征波段的提取,再用多元线性回归(MLR)、人工神经网络(ANN)和偏最小二乘回归(PLSR)进行建模。结果表明,水稻叶片氮素和籽粒蛋白质含量与特征光谱存在很好的模型关系,3种模型预测的决定系数(R2p)均在0.847以上,并以PLSR模型的预测效果为最好,可以实现水稻氮素营养和籽粒品质的高光谱估测。

Non-destructive and rapid monitoring methods for leaf nitrogen and seed protein of rice is of great significance in estimating rice nutritional diagnosis and grain quality monitoring, enhancing nitrogen management and use efficiency. In this study, a pot experiment was conducted to determine leaf nitrogen, grain protein content and canopy spectra in rice, and establish model to predict the leaf nitrogen and seed protein. Key spectral bands were selected by principal component analysis （PCA） method, and the predicted models were built by multiple linear regression （MLR）, artificial neural network （ANN） and partial least squares regression （PLSR） model. Results showed that significant correlation was found between leaf nitrogen and grain protein content and key spectral bands, and the determined coefficient of predication （R2p） were higher than 0. 847, and tested results predicted by PLSR models is the best. Therefore, it is concluded that rice nutrition and grain quality could be estimated by hyper spectral data.