三江源东部地区天然混合牧草营养成分近红外光谱预测模型构建

Development of Near-Infrared Spectroscopy Prediction Model for Nutritional Composition of Natural Mixed Forage in Eastern Region of Three-Rivers Source

本试验旨在构建三江源东部地区天然混合牧草营养成分近红外光谱预测模型。从三江源东部地区4个样点采集天然混合牧草样品301份,按照7∶3的比例将样品分为定标集和验证集。采用定标集样品的实测值及近红外光谱值进行粗蛋白质(CP)、中性洗涤纤维(NDF)和酸性洗涤纤维(ADF)预测模型的构建及交叉验证,采用验证集样品进行外部验证,进一步评价预测模型的预测效果。结果表明:CP预测模型构建中数据预处理的最佳方法为标准正态化处理+去散射处理+二阶求导,NDF和ADF预测模型构建中数据预处理的最佳方法为多元离散校正+二阶求导。CP、NDF和ADF预测模型的定标决定系数(R_(cal)^(2))、交叉验证相关系数(1-VR)均高于0.900,预测模型对验证集样品的模型预测值与化学实测值之间差异不显著(P>0.05),且验证决定系数(R_(CV)^(2))均高于0.900,交叉验证相对标准差(RPD_(cal))和外部验证相对标准差(RPD_(CV))均大于3.00。这表明近红外光谱技术可用于天然混合牧草营养价值的评定,本试验构建的预测模型预测效果较好,可应用于实际生产。

This experiment was conducted to build the near-infrared spectroscopy prediction model for nutritional composition of natural mixed forage in eastern region of three-rivers source.A total of 301 natural mixed forage samples were collected from 4 sample points in the eastern region of the three-rivers source,and the samples were divided into calibration set and verification set according to the ratio of 7∶3.The measured values and near-infrared spectroscopy values of the calibration set samples were used to construct and cross-validate the crude protein(CP),neutral detergent fiber(NDF)and acid detergent fiber(ADF)prediction models,and the values of verification set samples were used for external verification to further evaluate the prediction effect of the prediction models.The results showed that the best method of data multiprocessing in the construction of CP prediction model was standard normalization processing+de-scattering processing+second-order derivation,and the best method of data multiprocessing in the construction of NDF and ADF prediction models was multivariate dispersion correction+second order derivation.The determination coefficient of calibration(R_(cal)^(2))and cross-validation correlation coefficient(1-VR)of CP,NDF and ADF prediction models were all higher than 0.900,and the model prediction value predicted by the models for the validation set samples had no significant difference with the chemical measured value(P > 0.05),and the determination coefficient of verification(R_(CV)^(2))was higher than 0.900,and the cross-verify relative standard deviation(RPD_(cal))and external validation relative to standard deviation(RPD_(CV))were both greater than 3.00.The results indicate that near-infrared spectroscopy technology can be used to evaluate the nutritional value of natural mixed forage,and the prediction models in this study have good prediction effects,it can be applied to the actual production.