基于中红外光谱和机器学习算法的牛奶中β-乳球蛋白快速检测方法  

A mid-infrared spectroscopy and machine learning algorithm-based method for rapidly detecting content of β-lactoglobulin in milk

作　　者：邹慧颖 王东薇 樊懿楷 刘维华 杨俊华 余文莉 SABEK Ahmed Abdalla Ahmed Ibrahim 张淑君 ZOU Huiying;WANG Dongwei;FAN Yikai;LIU Weihua;YANG Junhua;YU Wenli;SABEK Ahmed Abdalla Ahmed Ibrahim;ZHANG Shujun(College of Animal Science and Technology,College of Veterinary Medicine,Huazhong Agricultural University,Wuhan 430070,China;Ningxia Institute of Veterinary Drug and Feed Supervision,Yinchuan 750000,China;Shijiazhuang Tianquan Breeding Dairy Co.,Ltd.,Shijiazhuang 050061,China)

机构地区：[1]华中农业大学动物科学技术学院、动物医学院,武汉430070 [2]宁夏兽药饲料监察所,银川750000 [3]石家庄天泉良种奶牛有限公司,石家庄050061

出　　处：《华中农业大学学报》2025年第2期125-133,共9页Journal of Huazhong Agricultural University

基　　金：国家重点研发计划项目(2023YFD1300400);中央高校基本科研业务费专项(2662023DKPY001);石家庄市科技计划项目(221500182A)。

摘　　要：为建立一种可以快速、批量、高效检测中国荷斯坦牛牛奶中β-乳球蛋白含量的方法,采集501份来自西北、华北和华中主要产奶地区的健康中国荷斯坦牛牛奶样本,采用高效液相色谱法测定牛奶样本中β-乳球蛋白的含量,并同步测定和收集牛奶样本中红外光谱数据(mid-infrared spectroscopy,MIRS)。以MIRS为预测变量,β-乳球蛋白含量为因变量,将12种光谱预处理方法进行连续2次的随机组合,并手动选取特征波段,使用偏最小二乘回归(partial least squares regression,PLSR)作为传统机器学习算法,建立预测牛奶中β-乳球蛋白含量的最优预测模型。结果显示:该模型交叉验证集和测试集的RC2和RP2分别为0.812 9、0.768 8,均方根误差RMSEC和RMSEP分别为0.476 2、0.524 9 g/L,性能偏差比(ratio of performance to deviation,RPD)为2.076 6,达到畜禽生产性能的测定要求。试验结果表明,可以利用MIRS建立模型预测中国荷斯坦牛牛奶中的β-乳球蛋白含量。501 milk samples of healthy Chinese Holstein cows were collected from major milk-producing regions in Northwest,North,and Central China to establish a method that can rapidly,in batch,and efficiently detect the content of β-lactoglobulin in milk from Chinese Holstein cows,high-performance liquid chromatography(HPLC) was used to determine the content of β-lactoglobulin in milk samples,and the mid-infrared spectroscopy(MIRS) data of milk samples were synchronously measured and collected 12methods of spectra pretreatment were randomly combined twice in a row,and the characteristic bands were manually selected with MIRS as the predictor variable and the content of β-lactoglobulin as the dependent variable.Partial least squares regression(PLSR) was used as a traditional machine learning algorithm to establish an optimal model for the prediction of the content of β-lactoglobulin in milk.The results showed that the RC2 and RP2 of the cross validation set and test set in the established model was 0.812 9 and 0.768 8,with the root mean square errors,RMSEC and RMSEP of 0.476 2 g/L and 0.524 9 g/L,the RPD of 2.076 6,meeting the requirements for measuring the production performance of livestock and poultry.It is indicated that MIRS can be used to establish a model for predicting the content of β-lactoglobulin in milk from Chinese Holstein cows.

关 键 词：中红外光谱 牛奶 Β-乳球蛋白 机器学习算法 光谱预处理 

分 类 号：S823[农业科学—畜牧学]