Rapid Discriminant Analysis Of The Distinction Of Liquor Brands Based On FTNIR Spectroscopy Combined With Chemometrics

China is the big country of the production and consumption of liquor. The distinction of well-known liquor brands is an important item on the quality testing of alcoholic products. The traditional analysis methods require chemical reagents, high professional skills, which are complex and not suitable for large-scale quality testing. In this study, based on the Fourier transform near infrared(FTNIR) spectroscopy, by combining principal component linear discriminant analysis(PCA-LDA) and partial least squares discriminant analysis(PLS-DA) methods with moving-window(MW) waveband selection mode, two integrated optimization methods for spectral pattern recognition were proposed, namely MW-PCA-LDA and MW-PLS-DA, which were successfully applied to the rapid discriminant analysis of the well-known liquor brands of Luzhou Laojiao.A total of 360 liquor samples were collected, comprising the 200 non-Luzhou Laojiao samples(positive) and 160 Luzhou Laojiao samples(negative) for comparison, and all liquor samples were in the same type of concentrated flavor and 52%vol. The FTNIR spectrum of the samples were collected by using the sample cells of 1 mm, 2 mm, 5 mm, 10 mm optical distance, respectively. In order to establish stable and reliable model, the whole samples were randomly divided into modeling and validation sets, then the modeling set was randomly divided 30 times into calibration, prediction sets. And the model parameters were selected based on the optimal prediction recognition rates(P_RECAve) of the 30 different divisions.Firstly, the models were established in the full spectrum band(15000-4000 cm-1). Using PCA-LDA method, the optimal prediction rate(P_RECAve) was 98.1%. The corresponding optimum distances were 2mm and 10 mm, respectively. Using PLS-DA method, the optimal P_RECAve was 100.0%. The corresponding optimum distance was 2mm. However, the number of wavenumber adopted reached up to 2852. So it is necessary to extract effective wavenumber and reduce model complexity. Secondly, further waveband optimization was performed based on the spectral data corresponding to the best optimum distance. Using MW-PCA-LDA method, the optimal model’s optimum distance was 2mm, the waveband was 5235-5130 cm-1, the value of N was 28, and the P_RECAve was 100.0%. Using MW-PLS-DA method, the optimal model’s optimum distance were 2mm and 10 mm, the wavebands were 5238-5204 cm-1 and 7186-7128 cm-1, the values of N were 10 and 16, the P_RECAve were both 100.0%. Finally, the randomly selected validation samples excluded in the modeling optimization process were used to validate the three optimal MW–PCA–LDA and MW-PLS-DA models. For 2 mm optical distance, the negative, positive recognition rates(V_REC—、V_REC+) of the optimal MW-PCA-LDA and MW-PLS-DA models were both 100%. For 10 mm optical distance, the values of V_REC—、V_REC+ of the optimal MW-PLS-DA model were 98.3%, 100.0%, respectively.The results indicated that FTNIR Spectroscopy combined with two integrated optimization methods MW-PCA-LDA and MW-PLS-DA can be applied for high-precision discrimination analysis of the distinction of liquor brands. The proposed waveband selection method could extract effective wavelengths and reduce model complexity, which provided valuable reference for designing a small dedicated spectrometer. The analytical methods proposed in this study were simple, rapid and effective, which had important application prospect for the in testing of liquor product quality in our country.