利用近红外光谱技术快速预测苜蓿干草营养成分含量

Rapid prediction of nutrient content of alfalfa hay by using near infrared spectroscopy

本研究旨在利用近红外光谱技术(near-infrared reflectance spectroscopy,NIRS)建立苜蓿(Medicago sativa)干草6种营养成分的近红外预测模型。分别从甘肃、宁夏、河北、江苏和陕西五省采集200份苜蓿干草样品,测定干物质(dry matter,DM)、粗灰分(Ash)、粗蛋白(crude protein,CP)、中性洗涤纤维(neutral detergent fiber,NDF)、酸性洗涤纤维(acid detergent fiber,ADF)和粗脂肪(ether extract,EE)的含量。选取苜蓿干草样品160份作定标集,40份作验证集。利用NIRS结合改良偏最小二乘法(modified partial least squares,MPLS)构建并验证其建立预测模型的优劣。结果表明:苜蓿干草DM、NDF含量预测模型的预测决定系数(coefficient of determination for validation,RSQ)和外部验证相对分析误差(ratio of performance to deviation for validation,RPD)分别为0.87和2.67、0.90和3.16,构建的模型可用于实际生产中的预测;CP、ADF含量预测模型的RSQ和RPD分别为0.83和2.41、0.82和2.28,构建的预测模型不能完全代替湿化学分析,但可用于大量样品的筛选分析;Ash含量预测模型的RSQ和RPD为0.59和1.51,构建的预测模型只能用于粗略的分析;EE含量预测模型的RSQ和RPD为0.45和1.32,构建的预测模型相关性较差,还需进一步优化。

The aim of this study was to establish a near-infrared prediction model for six nutrients of alfalfa hay using nearinfrared reflectance spectroscopy(NIRS).In total,200 samples of alfalfa hay were collected from five provinces(Gansu,Ningxia,Hebei,Jiangsu,and Shanxi)to analyze dry matter(DM),crude ash(Ash),crude protein(CP),neutral detergent fiber(NDF),acid detergent fiber(ADF),and ether extract(EE).The calibration and validation sets included 160 and 40 samples,respectively.The NIRS system was combined with modified partial least squares(MPLS)to construct and verify the prediction models.The results showed that the coefficient of determination for validation(RSQ)and ratio of performance to deviation for validation(RPD)of the DM prediction model of alfalfa hay were 0.87 and 2.67 and those of the NDF prediction model were 0.90 and 3.16,respectively.The constructed model can be used for prediction in actual production.The RSQ and RPD of the CP prediction model were 0.832.41 and those of the ADF prediction model were 0.82 and 2.28,respectively.The constructed prediction model cannot completely replace wet chemical analysis but can be used for the screening analysis of a large number of samples.The RSQ and RPD of the Ash content prediction model were 0.59 and 1.51,respectively.The constructed prediction model can only be used for a rough analysis.The RSQ and RPD of the EE content prediction model were 0.45 and 1.32,respectively.This constructed prediction model was poorly correlated and needed further optimization.