9种常用饲料原料生长猪有效能近红外定标模型的构建

Establishment of Near⁃Infrared Calibration Models for Predicting Available Energy in 9 Commonly Used Feed Ingredients in Growing Pigs

本试验旨在建立全脂米糠、大麦、麦麸、玉米蛋白粉、花生粕、棉籽粕、菜籽粕、次粉和玉米胚芽粕9种常用饲料原料的生长猪消化能和代谢能近红外定标模型,并探讨光谱范围和光谱预处理方法对定标效果的影响。试验分别选取了369个全脂米糠、521个大麦、174个麦麸、223个玉米蛋白粉、326个花生粕、237个棉籽粕、283个菜籽粕、138个次粉和160个玉米胚芽粕样品的光谱建立近红外定标模型。使用布鲁克MATRIX-I近红外光谱仪采集样品光谱,光谱采集范围为12000~4000 cm-1。采用偏最小二乘法建立近红外定标模型,利用动物试验测定值对所建模型进行外部验证。结果表明:有效能模型的最佳维数分布在7~14,最普遍的光谱预处理方法是一阶导数+矢量归一化(1st D+SNV)和一阶导数+多元散射校正(1st D+MSC)。此外,大麦猪消化能和代谢能定标模型的最佳光谱预处理方法是MSC,棉籽粕猪消化能定标模型的最佳光谱预处理方法是二阶导数(2nd D)。9种猪饲料原料的最佳消化能和代谢能近红外定标模型均取得了良好的定标效果,定标决定系数(R^(2)c)均在0.90以上,定标标准差(RMSEC)均在0.15以下,交互验证决定系数(R^(2)cv)均在0.90以上;除次粉猪消化能模型的交互验证标准差(RMSECV)为0.51外,其余有效能模型的RMSECV均在0.17以下;玉米蛋白粉猪代谢能模型的交互验证相对分析误差(RPDcv)最低,为3.42,花生粕猪消化能模型的RPDcv最高,达到了10.00。9种猪饲料原料最佳有效能模型的外部验证决定系数(R^(2)v)在0.62~0.92,外部验证标准差(RMSEP)在0.18~1.30,外部验证相对分析误差(RPDv)在1.11~2.80。因此,利用近红外反射光谱快速检测常用饲料原料生长猪的消化能和代谢能是可行的。本研究为把握猪饲料原料有效能的变异奠定了基础,对提高饲料利用效率以及实现猪的精准营养具有重要意义。

The aim of this study was to establish near⁃infrared calibration models of the digestible energy and metabolizable energy of 9 commonly used feed ingredients in growing pigs,including full⁃fat rice bran,bar⁃ley,wheat bran,corn gluten meal,peanut meal,cottonseed meal,rapeseed meal,wheat shorts,and corn germ meal,and to explore the effects of spectral regions and spectral pretreatment methods on the calibration performance.The near⁃infrared calibration models were established base on the spectra of 369 full⁃fat rice bran,521 barley,174 wheat bran,223 corn gluten meal,326 peanut meal,237 cottonseed meal,283 rapeseed meal,138 wheat shorts,and 160 corn germ meal,respectively.MATRIX⁃I near⁃infrared spectrometer was used to collect the spectrum.The wavenumber range was in the full spectra from 12000 to 4000 cm-1.All near⁃infrared calibration models were established using partial least squares(PLS)method.The external valida⁃tions of models were performed by comparing predicted values and animal experimental values.Results showed that the best factors of available energy models were from 7 to 14.The most common spectral pretreatment methods were first derivative+vector normalization(1st D+SNV)and first derivative+multiplicative scatter correction(1st D+MSC).In addition,MSC was the best spectral pretreatment method for the calibration mod⁃els of barley digestible energy and metabolizable energy and second derivative(2nd D)was the best spectral pretreatment method for the calibration model of cottonseed meal digestible energy.The best near⁃infrared cali⁃bration models of digestible energy and metabolizable energy of 9 feed ingredients showed good calibration per⁃formance,with coefficient of determination of calibration(R^(2) c)above 0.90,root mean square error of cali⁃bration(RMSEC)below 0.15,and coefficient of determination of cross⁃external validation(R^(2) cv)above 0.90.The root mean square error of cross⁃external validation(RMSECV)of all available energy models was below 0.17,except that the RMSECV of the digestible energy model of wheat shorts was 0.51.The residual predictive deviation of cross⁃external validation(RPDcv)of metabolizable energy model of corn gluten meal was the lowest(3.42),while digestible energy model of peanut meal was the highest(10.00).The coefficient of determination of external validation(R^(2) v),root mean square error of external validation(RMSEP),and residual predictive deviation of external validation(RPDv)of the best available energy models of 9 feed ingre⁃dients were 0.62 to 0.92,0.18 to 1.30,and 1.11 to 2.80,respectively.Therefore,it is feasible to rapidly de⁃tect the content of digestible energy and metabolizable energy in feed ingredients for growing pigs using near⁃infrared reflectance spectroscopy.This study lays a foundation for grasping the variation of available energy of feed ingredients in growing pigs and is of great significance for improving feed utilization efficiency and achie⁃ving accurate nutrition of pigs.