小麦籽粒蛋白质含量高光谱遥感预测模型比较  

Comparison of hyperspectral remote sensing prediction models for wheat grain protein content

作　　者：张志勇[1] 樊泽华 张娟娟[2] 邱青彬 郑亮 芮战许 马新明[2] 熊淑萍[1] ZHANG Zhiyong;FAN Zehua;ZHANG Juanjuan;QIU Qingbin;ZHENG Liang;RUI Zhanxu;MA Xinming;XIONG Shuping(College of Agronomy/Key Laboratory of Crop Growth and Development Regulation,Ministry of Education,Henan Agricultural University,Zhengzhou 450046,China;College of Information Management and Science,Henan Agricultural University,Zhengzhou 450046,China)

机构地区：[1]河南农业大学农学院/作物生长发育调控教育部重点实验室,河南郑州450046 [2]河南农业大学信息管理与科学学院,河南郑州450046

出　　处：《河南农业大学学报》2022年第2期188-198,共11页Journal of Henan Agricultural University

基　　金：国家自然科学基金项目(32071956);河南省重点研发与推广专项(212102110048);河南省高等学校重点科研项目(21A210015);国家重点研发计划(2021YFD1700900);河南省现代农业产业技术体系项目(S2010-01-G04)。

摘　　要：【目的】利用高光谱遥感技术实现冬小麦籽粒蛋白质含量的精准预测,比较筛选小麦籽粒蛋白质含量预测模型,实现优质小麦栽培生产。【方法】设置不同品质类型小麦品种和施氮量处理,测定开花期叶片叶绿素含量(SPAD)、叶片干物质质量(LDW)、地上生物量(AGB)、叶片氮含量(LNC)、叶片氮积累量(LNA)、叶面积指数(LAI)、植株氮含量(PNC)、植株氮积累量(PNA)和氮营养指数(NNI)9个农学参数及小麦冠层光谱,通过一阶导数和偏最小二乘法,构建基于不同农学参数的小麦籽粒蛋白质含量高光谱预测模型。【结果】一阶导数处理可以提高光谱数据与农学参数的相关性。运用偏最小二乘法构建的高光谱农学参数估测模型中以SPAD的模型建模精度与验证精度相对较优,建模集决定系数R^(2)与预测集标准均方根误差nRMSE分别为0.99和4.10%;NNI反演模型验证结果较好,相对预测偏差RPD为2.04;利用线性回归构建的农学参数-籽粒蛋白质预测模型中以LNC的建模精度与验证精度最佳,其建模集R^(2)、预测集均方根误差RMSE和RPD分别为0.64、0.79和2.11。最终构建的“高光谱-农学参数-籽粒蛋白质含量”预测模型以开花期LNC为中间变量的模型最优,其预测集R^(2)、RMSE和RPD分别为0.55、1.12和1.49。【结论】以农学参数为中间变量可以进行冬小麦籽粒蛋白质含量预测,“高光谱-LNC-籽粒蛋白质含量”具有较高精度的预测结果。【Objective】Hyperspectral remote sensing technology was used to accurately predict grain protein content of winter wheat,and the prediction models for grain protein content of wheat were compared and screened to realize the cultivation and production of high-quality wheat.【Method】Treatment of wheat varieties of different quality types and different nitrogen application rate were set up to determine wheat canopy spectra and nine agronomic parameters in the flowering period,including leaf chlorophyll content(SPAD),leaf dry weight(LDW),aboveground biomass(AGB),leaf nitrogen content(LNC),leaf nitrogen accumulation(LNA),leaf area index(LAI),plant nitrogen content(PNC),plant nitrogen accumulation(PNA)and nitrogen nutrition index(NNI).The hyperspectral prediction model for wheat grain protein content based on different agronomic parameters was constructed by first-order derivative and partial least squares methods.【Result】The results showed that the correlation between spectral data and agronomic parameters was improved by the first-order derivative processing.Among the hyperspectral agronomic parameter estimation models constructed by partial least square method,the modeling accuracy and verification accuracy of SPAD model were relatively excellent,and the determination coefficient R^(2) of the modeling set and the nRMSE of the prediction set were 0.99 and 4.10%respectively.The verification result of NNI inversion model was better,and the relative prediction deviation RPD reached 2.04.In the agronomic parameter-grain protein prediction model constructed by linear regression method,LNC had the best modeling accuracy and verification accuracy.The RMSE and RPD of modeling set R^(2) and prediction set were 0.64,0.79 and 2.11,respectively.The ultimately constructed prediction model of“hyperspectral-agronomic parameters-grain protein content”was the best model with flowering LNC as the intermediate variable,and its prediction sets R^(2),RMSE and RPD were 0.55,1.12 and 1.49,respectively.【Conclusion】The grai

关 键 词：冬小麦 籽粒蛋白质含量 农学参数 高光谱遥感 预测模型 

分 类 号：S512.1[农业科学—作物学]