近红外光谱的预处理对羊肉TVB-N模型的影响

为研究能否通过对算法参数的调整和算法的组合来减弱甚至消除便携式近红外仪和样品组织结构等对样品光谱信息的影响,提高模型的预测准确性和稳健性,实现现场快速无损检测生鲜羊肉挥发性盐基氮(total volatile basic nitrogen,TVB-N)的目的。本研究应用不同参数组合的单一算法和不同算法组合对样品的光谱信息进行预处理并建模,从模型的预测准确性和稳健性2个方面探讨算法参数和算法组合对模型性能的影响,找出针对检测生鲜羊肉中TVB-N含量的最佳预处理方法。结果表明,不同的算法参数和算法组合对模型性能的影响差别很大,对样品的近红外光谱信息进行差分求导(窗口数为6,求导阶次为1)后,模型性能最佳。模型的校正标准差和验证标准差分别为1.21和1.31,校正标准差和验证标准差的比值为1.08小于1.2,主成分数为10,校正集相关系数和验证及相关系数分别为0.94和0.92。说明通过对算法参数的调整和对算法的组合可以有效提高模型性能,满足应用便携式近红外仪现场快速无损检测生鲜羊肉TVB-N含量的要求。

Effects of Spectral Pretreatments on Prediction of Total Volatile Basic Nitrogen (TVB-N) Content in Mutton Using Near Infrared Spectroscopy

This study aimed at in situ, rapid and nondestructive detection of total volatile basic nitrogen (TVB-N) content in fresh raw mutton using near infrared spectroscopy. We checked whether the impact of portable near infrared spectrometer and microstructure of samples on the spectral information of the samples could be reduced or even eliminated by adjusting algorithm parameters and combing different algorithms for the purpose of improving the accuracy and robustness of the prediction model developed. Various individual algorithms with different parameter combinations and various algorithm combinations were used to pretreat the spectral information of the samples for modeling. The effects of algorithm parameters and algorithm combinations on the performance of the model in terms of predictive accuracy and stability were evaluated and discussed to find the optimal pretreatment method. The results showed that different algorithm parameter combinations and different algorithm combinations distinctly affected the model performance. When the spectral information of the sample was pretreated with difference derivatives (window parameter is 6, and order of differentiation is 1), the best model performance was achieved. The standard error of calibration (SEC) and standard error of prediction (SEP) of the model were 1.21 and 1.31, respectively, with SEP/SEC = 1.08 < 1.2. The number of principal components was 10. The correlation coefficients of calibration and prediction were 0.94 and 0.92, respectively. Our study verified that spectral information pretreatment with proper algorithm parameter combination and algorithm combination can significantly improve the model performance and allow fast, non-destructive and on-the-spot detection of TVB-N in mutton.