Abstract: Abstract: In this study, a prediction model based on Fourier transform mid-infrared spectroscopy（FTIR） was developed for the rapid detection of β-lactoglobulin in milk. The content of β-lactoglobulin in 260 different batches of milk samples was determined quantitatively by high-performance liquid chromatography-mass spectrometry, and the mid-infrared spectra of raw milk were collected by FT1 milk composition analyzer, the background noise was eliminated by Savitsky-Golay (SG) smoothing, first-order derivative or second-order derivative pretreatment, and then the β-lactoglobulin prediction model was constructed by using partial least squares regression (PLSR) method. Results: the β-lactoglobulin prediction model was established by comparing different mid-infrared spectral preprocessing methods and calibrating the preprocessing spectra after the final selection number and smoothing the baseline shift. The results showed that the standard normal transformation method was the optimum preprocessing method,the correlation coefficient R2 of the calibration set was 0.943, the root mean square error of calibration ( RMSEC) value was 0.049% ,the correlation coefficient R2 of the test set was 0. 923 and the root mean square error of prediction ( RMSEP) value was 0.057% .

Key words: FTIR, β-lactoglobulin, milk, partial least squares

摘要： 通过建立基于傅里叶变换中红外光谱预测模型实现牛乳中β-乳球蛋白的快速检测。将收集到的260个不同批次的牛乳样品使用高效液相色谱定量检测β-乳球蛋白含量作为参比值；通过FT1乳成分分析仪采集牛乳样品的中红外光谱，选择有效波段，并将原始中红外光谱先经过Savitsky-Golay（SG）平滑、一阶导数或二阶导数等预处理方法消除背景噪音，再利用偏最小二乘回归（ Partial Least Squares Regression,PLSR）法构建β-乳球蛋白预测模型。通过比较不同中红外光谱预处理方法，最终选择Savitzky-Golay 5点平滑二阶导数的最小二乘法（ Partial Least Squares,PLS）模型为最优模型，获得校正集相关性系数 R2为0.943，校正均方差( Root Mean Square Error of Calibration,RMSEC) 为 0.049% ，预测集相关性系数 R2 为 0.923，预测均方差(Root Mean Square Error of Prediction, RMSEP) 为 0.057% ，具有良好的准确性。

关键词: 傅里叶变换中红外光谱技术, β-乳球蛋白, 牛乳, 偏最小二乘回归法