基于近红外光谱技术快速检测广式酱油发酵过程中主要理化指标的含量

Rapid Determination of Main Physicochemical Indexes of Cantonese Soy Sauce in the Fermentation Process Based on Near-infrared Spectroscopy

摘要:

为了实现对广式酱油中理化指标含量的快速检测，该研究应用近红外光谱技术（Near Ifrared Spectroscopy, NIRS）结合5种不同光谱预处理方法以及2种特征波段筛选方法，运用偏最小二乘回归法与支持向量回归法建立了广式酱油中氨基酸态氮、总氮、可溶性无盐固形物、总酸、铵盐、总糖、还原糖以及盐分8种理化指标的定量预测模型，并比较建模效果，以筛选最佳定量预测模型。结果表明，支持向量回归法所建模型的决定系数均高于PLSR模型，均方根误差均低于PLSR模型，说明采用支持向量回归建模效果更优。结果表明，经异常样品剔除、预处理、特征波段筛选后，针对广式酱油8种理化指标的建立的支持向量回归定量预测模型效果较好，其中，各理化指标定量预测模型的训练集决定系数R2为0.991 1~0.962 1，测试集决定系数R2为0.977 9~0.857 9。同时对各理化指标筛选出的最佳定量预测模型进行外部验证，结果显示，广式酱油各指标的预测值与化学值之间的绝对误差≤1.31，t检验结果显示两组数据的无显著性差异，所建立的定量预测模型能够快速准确检测广式酱油中各理化指标的含量。该研究可以快速无损测定广式酱油发酵过程的8个重要理化指标，为酱油实际工业生产过程中的质量控制奠定基础。

Abstract:

The production of soy sauce is crucial in the food industry. This study aimed to achieve the rapid detection of eight physicochemical indexes of Cantonese soy sauce, namely the content of amino acid nitrogen, total nitrogen, soluble salt-free solids, total acid, ammonium salt, total sugar, reducing sugar, and salt. To achieve this goal, first, quantitative prediction models were established for these physicochemical indexes using near-infrared spectroscopy (NIRS). Specifically, the NIRS spectra of Cantonese soy sauce were processed with five spectral preprocessing methods and two feature band screening methods, and the processed spectra were used to construct quantitative prediction models through partial leastsquares regression (PLSR) and support vector regression (SVR). Next, the performances of the models were compared to screen the optimal quantitative prediction models. The results showed that compared with the PLSR-based quantitative prediction models, the SVR-based quantitative prediction models had higher coefficients of determination (R2) and lower root-mean-square errors, indicating that the SVR-based quantitative prediction models were superior after abnormal sample removal, preprocessing, and feature band screening. The R2 values of the training and test sets for each physicochemical index were between 0.991 1~0.962 1 and 0.977 9~0.857 9, respectively. Furthermore, the optimal quantitative prediction models screened for each physicochemical index was externally validated. The absolute error between the predicted and chemical values of each index of the Cantonese soy sauce was ≤ 1.31, and the t-test results showed that there was no significant difference between the two groups of data, indicating that the quantitative prediction models can quickly and accurately detect each physicochemical index of the Cantonese soy sauce. The proposed models can quickly and nondestructively determine eight important physicochemical indexes of Cantonese soy sauce in the fermentation process, laying a foundation for quality control in the actual industrial production process of soy sauce.