Preliminary Assessment of Parmigiano Reggiano Authenticity by Handheld Raman Spectroscopy

Abstract

Raman spectroscopy, and handheld spectrometers in particular, are gaining increasing attention in food quality control as a fast, portable, non-destructive technique. Furthermore, this technology also allows for measuring the intact sample through the packaging and, with respect to near infrared spectroscopy, it is not affected by the water content of the samples. In this work, we evaluate the potential of the methodology to model, by multivariate data analysis, the authenticity of Parmigiano Reggiano cheese, which is one of the most well-known and appreciated hard cheeses worldwide, with protected denomination of origin (PDO). On the other hand, it is also highly subject to counterfeiting. In particular, it is critical to assess the authenticity of grated cheese, to which, under strictly specified conditions, the PDO is extended. To this aim, it would be highly valuable to develop an authenticity model based on a fast, non-destructive technique. In this work, we present preliminary results obtained by a handheld Raman spectrometer and class-modeling (Soft Independent Modeling of Class Analogy, SIMCA), which are extremely promising, showing sensitivity and specificity of 100% for the test set. Moreover, another salient issue, namely the percentage of rind in grated cheese, was addressed by developing a multivariate calibration model based on Raman spectra. It was possible to obtain a prediction error around 5%, with 18% being the maximum content allowed by the production protocol.

Keywords: PLS; Parmigiano Reggiano; SIMCA; authenticity; chemometrics; grated cheese; handheld Raman.

Figure 1

Raw (top) and aligned spectra (bottom).

Figure 2

Aligned spectra after smoothing and baseline correction (top), normalization to unit area (bottom left) and probabilistic quotient normalization (bottom right).

Figure 3

One-class Soft Independent Modeling of Class Analogy (SIMCA) model built using authentic PR samples (PR1st and PR2nd, in Table 1) as a calibration set. (a) Sensitivity, specificity and efficiency in CV vs. number of components. Specificity was estimated on five not-PR samples. (b) Sensitivity in CV (red line, right y-axis) and RMSECV (blue line, left y-axis) vs. number of components. The shaded areas in each graph highlights the number of components selected according to each one of the criteria, Sensitivity_CV (red shade), Efficiency_CV (green shade) and RMSECV (blue shade), respectively. Sensitivity in (a) and (b) are the same.

Figure 4

Reduced score distance ($T^2$/$T^2{}_{lim}$) vs. reduced orthogonal distance (Q/Q_lim), at 95% confidence level. (a) One-class SIMCA model with two components, selected according to maximum Sensitivity_CV; (b) One-class SIMCA model with five components, selected according to maximum Efficiency_CV (or selected according to minimum RMSECV, i.e., the model is the same). Efficiency_CV has been estimated by the not-PR cal. samples (green triangles). The red circle (radius = 1.414) corresponds to the acceptance class boundary.

Figure 5

Partial least squares regression (PLS) calibration model for % rind content based on the first sampling set: Measured vs. Predicted rind values. PR 1st samples: black diamonds; PR 2nd set samples: red squares.

Figure 6

PLS calibration model for % rind content based on the updated calibration set (PR 1st + 6 PR 2nd): only predicted samples are shown, measured vs. predicted rind values. PR 2nd samples (test): red squares; unknown-PR: blue triangles. Samples falling in the green framed area are correctly seen as compliant and samples in the red frame are correctly seen as non-compliant.

Figure 7

PLS calibration model for % rind content based on the updated calibration set (PR 1st + 6 PR 2nd): regression coefficients with region with variable importance in projection (VIP) scores higher than one highlighted in red.

Figure 8

For the same sample the spectrum acquired on grated PR as such (red line) and on packaged grated PR (blue line) are shown. Where the plastic absorbs, the intensity of the blue line is higher than the intensity of the red one.