应用近红外技术快速预判生猪血液指标及劣质肉

Fast Prediction of Pig Blood Parameters Using Near-Infrared Spectroscopy for Discrimination of Inferior Quality Pork

为了实现白肌(pale soft exudative,PSE)肉和黑干(dark fi rm dry,DFD)肉等劣质猪肉的预判,实验收集了生猪屠宰时的血液样品,其中64个样品用于建立血液皮质醇浓度预测模型,89个样品用于建立血液葡萄糖浓度预测模型。应用便携式近红外仪采集样品的近红外光谱信息并使用不同算法和算法组合对样品的光谱信息进行预处理后利用偏最小二乘回归算法进行建模。通过模型评价参数对预处理方法进行筛选后发现,针对预测生猪血液中的葡萄糖浓度,对样品的近红外光谱信息进行Savitzky-Golay求导和基线校正后建模,模型性能最佳。模型的校正标准差和验证标准差分别为2.07和2.48,主因子数为6,校正集相关系数和验证集相关系数分别为0.88和0.85。针对预测生猪血液中的皮质醇浓度,对样品的近红外光谱信息进行标准化、差分求导、Savitzky-Golay平滑和净分析信号后建模,模型性能最佳。模型的校正标准差和验证标准差分别为0.05和0.15,主因子数为6,校正集相关系数和验证集相关系数分别为0.97和0.67。应用筛选出的模型对另外采集的未用于建模的25个生猪血液样品中的葡萄糖和皮质醇浓度进行检测,从而进行劣质猪肉预警,PSE和DFD肉的预判准确率分别达到92%和96%。说明应用便携式近红外仪检测生猪血液中的葡萄糖和皮质醇浓度,从而预判劣质猪肉的方法是可行的。

In order to quickly discriminate pale, soft and exudative（PSE） pork and dark, fi rm, and dry（DFD） pork, 64 blood samples were collected at slaughter and tested by near-infrared（NIR） spectroscopy for the establishment of a predictive model for serum cortisol concentration, and 89 additional sample were collected and also detected by NIR spectroscopy for the development of a model to predict serum glucose concentration. Spectral information was acquired employing a portable NIR spectrometer and preprocessed using individual and combined algorithms for modeling using partial least square regression. Based on evaluation of the model parameters, the predictive model for serum glucose concentration developed by spectral pretreatment using Savitzky-Golay derivative ＋ baseline correction had the best performance. The standard error of calibration and standard error of prediction of the model were 2.07 and 2.48, respectively and the number of principal components was 6. The correlation coeffi cients of calibration and prediction sets were 0.88 and 0.85, respectively. The optimal spectral pretreatment method for predictive modeling of serum cortisol concentration was autoscaling ＋ difference derivative ＋ Savitzky-Golay smoothing ＋ net analyte signal. The standard error of calibration and standard error of prediction of the predictive model for serum cortisol concentration were 0.05 and 0.15, respectively and the number of principal components was 6. The correlation coeffi cients of calibration and prediction sets were 0.97 and 0.67, respectively. The built models were respectively used to predict serum glucose and cortisol concentrations of 25 unknown samples and consequently recognize PSE and DFD meat with an accuracy of 92% and 96%, respectively. Hence, it is feasible to predict serum glucose and cortisol concentrations of pig blood using NIR spectroscopy in order to identify PSE and DFD meat.