Miniaturized wavelength model optimization for visible–near-infrared spectroscopic discriminant analysis of soy sauce adulteration identification

Using visible and near-infrared (Vis–NIR) spectroscopy, the discriminative analysis models for a brewed soy sauce brand and its high imitation adulterated samples were established. The “blended soy sauce” was prepared with salt water, monosodium glutamate, and caramel color, and added to brewed soy sauce in different adulteration rates, formed adulterated soy sauce samples. Short- and long-measurement modals (1 mm & 10 mm cuvettes) were used to collect the transmission spectra, help to improve spectral quality and category characteristics in long- and short-wavelength regions. Based on partial least squares-discriminant analysis (PLS-DA), moving-window waveband screening and wavelength step-by-step phase-out, were used for wavelength optimization of the first and the second stages, respectively, jointly denoted as MW-WSP-PLS-DA. For 1 mm case, four optimal models ( N = 5, 5, 3 and 3) with MW-WSP-PLS-DA from visible (400–780 nm), short-NIR (780–1100 nm), and long-NIR (1100–2498 nm) regions were selected, respectively. Through independent validation, their total recognition-accuracy rates in validation (RAR V ) reached 95.2%, 99.1%, 100% and 100%, respectively. For 10 mm case, three optimal models ( N = 3, 3 and 7) from visible (400–780 nm) and NIR high-overtone frequency (780–1388 nm) regions were selected, respectively. Their RAR V values reached 97.8%, 99.1% and 97.8%, respectively. For short- and long-measurement modals (1 mm & 10 mm), RAR V values of the global optimal models of three wavelengths located exactly in long- and short-wavelength regions, reached 100% and 99.1%, respectively. It indicated that Vis–NIR spectroscopy and miniaturized wavelength strategy with MW-WSP-PLS-DA could be used for high-precision soy sauce adulteration identification. It provided a valuable reference for designing dedicated and miniaturized spectrometers for different measurement modals and different spectral regions.