Through the bottle authentication of red wine using near-IR fluorescence spectroscopy

TL;DR

The paper tackles non-invasive authentication of unopened red wine bottles by introducing an axicon-based through-bottle fluorescence system that uses a single 785 nm excitation to elicit wine autofluorescence while suppressing the bottle signal. Spectral data are analyzed with PCA to reveal varietal groupings and LDA to achieve high-accuracy classification, including 100% correct labeling of 20 wines and 96.7% accuracy when wines are measured through different bottles. The method provides a fast, non-destructive fingerprint that can enable on-site wine authentication with a compact setup and holds potential for extension to other packaged high-value goods. This approach addresses a critical need in food safety and provenance verification by enabling direct analysis of contents without opening the container.

Abstract

A major unaddressed challenge for food science remains the accurate characterisation of contents in sealed containers with a non-invasive method. This issue is particularly pressing for tackling fraud in the red wine industry, valued at billions of dollars globally, where product authenticity, brand reputation, and consumer trust are paramount. Whilst many techniques exist for authenticating wine externally, to date performing accurate classification of the contents within unopened bottles remains elusive. Using only a single near-infrared optical excitation source operating at a wavelength of 785 nm, in combination with a bespoke geometry to circumvent the confounding signal of the glass, we demonstrate that through-bottle fluorescence spectra can distinguish between twenty different red wines in their original, intact bottles. All twenty wine bottles were correctly classified with linear discriminant analysis (LDA) and principal component analysis (PCA) revealed strong varietal grouping. This non-invasive and rapid technique has the potential to enable on-site, routine wine authentication to combat the growing issue of wine fraud. The geometry itself is applicable across multiple fields for the analysis of other high-value products through their packaging, where authenticity verification is critical.

Figures (11)

• Figure 1: Concept and characteristics of the through-bottle fluorescence system. a) The fluorescence excitation beam (785 nm) is shaped so that an annular beam forms on the surface of the bottle, after which it focuses to a point inside the bottle. The fluorescence emission from the wine, excited at this focal point, is then collected through the centre of the annular excitation beam to avoid the fluorescence signal from the glass bottle. The iris blocks any fluorescence signal from the glass that is excited by the incident annular beam. Lens 1 has a focal length of 40 mm. b) The fluorescence spectrum of red wine (Shiraz 1) acquired through the bottle; the fluorescence spectra of only the bottle and only the wine are also plotted for reference. c) The fluorescence spectra measured as the wine bottle (Shiraz 2) is moved along the beam path. The spectra were normalised for clarity, and the fluorescence spectra of only the bottle and only the wine are plotted as dashed lines for reference. The fluorescence signal from the bottle is suppressed as the focus moves into the bottle. The relative contributions of the wine and the bottle to the total fluorescence signal are presented in the bar graphs for d) different bottle positions, and e) different iris diameters. For this wine bottle, the optimal system parameters are 30 mm from Lens 1 and an iris size of 7 mm. The insets show the excitation beam profile incident on the surface of the bottle (i.e. after being focused by Lens 1) in grey and the collection region in red. Scale bars are 1 mm.
• Figure 2: Axicon and conventional Gaussian beam setup comparison. The fluorescence spectra of red wine (Grenache 1) were measured through a bottle with a) the axicon setup and b) a conventional Gaussian beam setup. Ten spectra (solid lines) were taken at different positions around a single bottle. The spectra obtained with the axicon system were consistent and dominated by the fluorescence signal from the wine, whereas the spectra taken with a Gaussian beam varied depending on the bottle thickness. The fluorescence spectra solely of the wine and solely of the bottle are added for reference (dashed lines).
• Figure 3: Transmission metric to determine the optimal excitation wavelength for the through-bottle fluorescence system. a) Transmission spectra of a typical red wine bottle and red wine (Shiraz). The insets show the propagation of a 532 nm and a 785 nm annular beam through red wine (Shiraz 2); 100 mW laser power used for excitation. For the through-bottle system to work, the excitation beam must be transmitted through the green glass and the wine to create a focal point inside the wine for fluorescence excitation. b) A plot of the transmission spectrum of the glass bottle $\times$ the transmission spectrum of the wine. The peak position can be used to identify the optimal excitation wavelength for the through-bottle setup; the chosen excitation wavelength of 785 nm is indicated on the plot.
• Figure 4: Principal component (PC) plot discriminating between twenty different unopened wines. This PC plot shows the grouping of 10 spectra of each wine, with the different varietals tending to group together. The first three principal components describe 99.4% of the variation. The inset on the right shows that the five Shiraz/Syrah varietals that clustered together can be further separated when considering the third PC.
• Figure 5: Classification of the fluorescence spectra of 6 wines acquired through different bottles. The LDA model was trained on the spectra measured through one of the bottles ($n_{train} = 60$). The confusion matrix shows the model performance on a test set consisting of the spectra measured through the five remaining bottles ($n_{test} = 300$). Only ten spectra were misclassified, giving a success rate of 96.7%.