Vis-NIR Spectroscopy Combined With Chemometric Methods Applied To Multi-Brand Identification And Adulteration Discriminant Analysis Of Soy Sauce

Soy sauce is one of the most widely used condiments.High-quality brewed soy sauce is produced by advanced microbial fermentation technology and quality control method.It is delicious and nutritious,and is deeply loved by consumers.The brand and adulteration identification of soy sauce can avoid adulteration and fraud,which is meaningful for food safety screening.The relevant analytical methods mainly include manual identification and chromatography methods.The former has obvious subjective bias and low efficiency;and the latter usually requires quantitative analysis of a variety of minute quantity characteristic components,which requires reagents,complex operations,time-consuming and high cost.Visible-near infrared（Vis-NIR）spectroscopy is a rapid and simple detection technique without reagents.In recent years,Vis-NIR spectroscopy has been used for quantitative analysis of soy sauce quality indicators（such as total acid,total nitrogen,and amino acid nitrogen,etc.）,but there are few applied studies in soy sauce brand identification and adulteration discrimination.In this paper,Vis-NIR spectroscopy combined with k-nearest neighbor（k NN）and partial least squares-discriminant analysis（PLS-DA）methods were used to carry out two aspects of research on the qualitative analysis of soy sauce:taking 3 soy sauce brands（discrimination）and 10 soy sauce brands（interference）as experimental objects,a four-category discriminant model of soy sauce brand was established;taking a certain brand of brewed soy sauce（negative）and adulterated soy sauce（positive）as the experimental objects,a two-category discriminant model of soy sauce adulteration was established.Considering that the use of long-and short-transmission measurement methods can highlight the absorption difference in the short-wavelength region and avoid the saturable absorption in the long-wavelength region,this paper uses short-and long-optical paths（1 mm,10 mm）to fully extract the characteristic spectral information of different categories of samples.The main research contents and results are as follows:1.Vis-NIR spectral analysis of multi-brand identification of soy sauce:Chubang,Haitian,and Mastar soy sauce were used as the identification brands（not in order as category-I,II,III）,and the other 10 brands of soy sauce were used as interference brands（category-IV）to establish the four-category discriminant model of soy sauce with 1 mm and 10 mm measurement modals,respectively.A rigorous calibration-prediction-validation sample experiment design was adopted.The calibration and prediction sets were used for modeling and parameter optimization;and the independent validation samples that not involved in modeling were used to validate the selected models,thereby obtained objective evaluation.Based on the k NN algorithm,combined with the moving window（MW）waveband screening and wavelength step-by-step phase-out（WSP）methods,the multi-stage discriminant analysis models of k NN,MW-k NN and MW-WSP-k NN were established respectively,and the efficient wavelengths with high accuracy and few wavelength models were selected.For the dataset of 1 mm measurement modal,35 dual-wavelength and one three-wavelength optimal MW-WSP-k NN models were determined.After independent validation,the total recognition accuracy rates(RAR_Total)all reached 100%.For the dataset of 10 mm measurement modal,7 three-wavelength optimal MW-WSP-k NN models were determined.After independent validation,the RAR_Total reached more than 96.8%,and the optimal RAR_Total was 97.8%.2.Vis-NIR spectral analysis of adulteration identification of soy sauce:a brand of brewed soy sauce was used as identification sample（negative）.According to the color and the main components’content（salt,amino acid nitrogen）of brewed soy sauce,the mother solution sample of“blended soy sauce”was concocted of Na CI solution,monosodium glutamate and caramel color.and then the adulteration soy sauce samples（positive）were prepared by mixing the brewed soy sauce and the blended soy sauce according to different proportions.The calibration-prediction-validation process was used to establish the discriminant analysis model for the above two categories of sample.Standard normal variable（SNV）was used for spectral preprocessing;PLS-DA combined with MW waveband screening and WSP methods were usede to establish MW-WSP-PLS-DA model,which was applied to two-category discriminant model of soy sauce adulteration.For the dataset of 1 mm measurement modal,2 three-wavelength optimal models were determined in the long-wavelength region,and their modeling and validation RAR_Totalreached 100%;for the dataset of the 10 mm measurement modal,2 three-wavelength optimal models were determined in the visible spectral region,their modeling effects reached 100%,and their validation RAR_Total were 97.9%,98.5%,respectively.The results showed that:（1）The multi-modal miniaturized Vis-NIR spectroscopy technology has the feasibility of high-precision multi-brand identification of soy sauce.Among them,in the 1mm and 10 mm measurement modals,only 2 and 3 wavelengths were used,respectively,and the discrimination accuracy can reach more than 96.8%;（2）The multi-modal miniaturized Vis-NIR spectroscopy technology also has the feasibility of high-precision identification of soy sauce adulteration.Among them,in the 1 mm measurement modal,only 3 wavelengths can be used to achieve a 100%discrimination accuracy;in the 10 mm measurement modal,only 3 wavelengths can be used to achieve a discrimination accuracy of more than 97.9%.This technology is rapid,simple,miniaturized and low-cost,and has great potential for field application,which is important for food safety and livelihood construction.