滩羊肉中油酸和亚油酸含量的近红外预测模型建立

基于近红外光谱技术结合化学计量学方法建立滩羊肉中油酸和亚油酸含量的预测模型。选取滩羊肉外脊、里脊、羊霖、羊腩共138 份样本，在900～2 500 nm波长范围内，采集滩羊肉糜样品的近红外反射光谱，利用气相色谱法作为参考，测定样品中油酸和亚油酸含量，并建立滩羊肉中油酸和亚油酸含量的偏最小二乘回归（partial least squares regression，PLSR）模型。为优化模型性能，使用间隔随机蛙跳（interval random frog，IRF）算法进行数据降维处理。结果表明：对于油酸模型，经过标准正态变量变换结合一阶导数处理后的全波长模型相关性较高，校正相关系数（Rc）和交叉验证均方根误差（root mean square error of cross-validation，RMSECV）分别为0.889 5和10.2515，预测相关系数（Rp）和预测集均方根误差（root mean square error of prediction set，RMSEP）分别为0.7357和10.2492，然而，经IRF算法提取92个特征波长后，Rc和Rp均低于全波长模型；对于亚油酸模型，使用多元散射校正处理后的全波长模型Rc最大，为0.8747，RMSECV为1.0512，但其Rp和RMSEP较小，利用IRF算法提取102 个特征波长后，建立的亚油酸模型相关性得到极大改善，其中Rc最大达到0.9912，相应的RMSECV为0.0118，Rp为0.9879，RMSEP为0.0122。因此，近红外光谱技术结合IRF算法不能较好预测滩羊肉中油酸含量，但对亚油酸含量具有较好的预测能力。

Establishment of Predictive Models for Determination of Oleic Acid and Linoleic Acid Contents in Tan Sheep Meat by Near Infrared Spectroscopy

Predictive modelling was performed to determine oleic acid and linoleic acid contents in Tan sheep meat using near infrared spectroscopy combined with chemometrics. In this study, 138 samples of striploin, tenderloin, thick flank and belly from Tan sheep were collected. Near infrared reflectance spectra of minced meat samples were collected within the wavelength range of 900–2 500 nm. The contents of oleic acid and linoleic acid in the samples were determined by gas chromatography (GC) and the analytical data obtained were used as a reference to establish a partial least squares regression (PLSR) model. The performance of the model was optimized by reducing the dimension of the data using the interval random frog (IRF) algorithm. The results showed that the full-wavelength model for oleic acid developed through spectral preprocessing using standard normal variate (SNV) transformation combined with first derivative exhibited a higher correction coefficient of calibration (Rc) of 0.889 5 compared with any other spectral preprocessing method, with root mean square error of cross-validation (RMSECV) of 10.251 5, correlation coefficient of prediction (Rp) of 0.735 7, and root mean square error of prediction set (RMSEP) of 10.249 2. The Rc and Rp were both higher than those based on 92 characteristic wavelengths extracted using IRF. For linoleic acid, the full-wavelength model established with multiplicative scatter correction (MSC) presented the highest Rc of 0.874 7 with RMSECV of 1.051 2 but the Rp and RMSEP were lower. The model based on 102 characteristic wavelengths extracted by IRF showed greatly improved correlation, with the highest Rc of 0.991 2, as well as RMSECV of 0.011 8, Rp of 0.987 9, and RMSEP of 0.012 2. Therefore, near-infrared spectroscopy combined with IRF algorithm cannot well predict the oleic acid content of Tan sheep meat, but it has high predictive ability for the linoleic acid content.