Optical Fiber-surface plasmon resonance (SPR) sensors with machine learning-assisted for label-free identification of animal sources-gelatine

Abstract

Label-free optical techniques such as surface plasmon resonance (SPR) are attractive for food authentication because they combine high refractive-index sensitivity with real-time, reagent-free operation. Discriminating gelatine by animal source remains difficult, however, because the refractive indices of bovine, porcine, and fish gelatine solutions at a matched concentration differ by only 1.4 × 10⁻⁴ to 2.1 × 10⁻³ refractive index units (RIU), an order of magnitude less than the change produced by concentration alone. This work reports a reflection-mode fiber-optic SPR (FO-SPR) probe, formed by sputtering a 50 nm gold film onto the unclad core of a 400 μm multimode fiber and mirroring the distal end face with silver paste, combined with a full-spectrum machine-learning (ML) classifier. Refractive indices of gelatine solutions from 1% to 8% (w/v) were measured by refractometry and correlated with the measured resonance-wavelength shift of the probe. The sensor exhibits refractive-index sensitivities of ≈ 1719, 1749, and 1316 nm/RIU for porcine, bovine, and fish gelatine, respectively. Whole SPR spectra (350–1100 nm) were smoothed with a Savitzky–Golay filter and projected onto principal components, the first two of which carry 98.09% of the spectral variance and classified with a support vector machine using a radial basis function kernel. Smoothing, scaling, and the principal-component basis are fitted on training data only, and generalization is assessed with replicate-grouped and leave-one-concentration-out cross-validation, so that the classifier is tested for source specificity rather than concentration tracking. The classifier reaches an overall accuracy of 96.55% on held-out spectra, with per-class F1 scores between 0.917 and 1.000, whereas a classifier restricted to the scalar refractive index makes its errors systematically between the bovine and fish classes whose refractive indices overlap most closely. FTIR spectra of the same materials provide chemical corroboration. The study is a proof of concept for source discrimination among pure, laboratory-prepared gelatine solutions; the mixture, matrix, and inter-batch experiments required before the platform can support routine halal verification are stated explicitly.