Raman and NIR spectroscopy: A discussion of calibration robustness for food quality measurements through two case...

Fast and nondestructive spectroscopic measurements of food quality are of great interest to the food industry due to their huge potential in decision support and optimization of processes and food value chains. Choosing the type of spectroscopic technique to invest in is not always straight-forward. In this work, we discuss the robustness aspect of Raman and NIR spectroscopy as a relevant factor to consider. We hypothesize that Raman spectroscopy has an advantage in calibration development and maintenance due to the higher degree of chemical details in the spectra, lower degree of band overlap and lower sensitivity to water.

Robustness was discussed through two case studies, assessing (1) the transferability of intact strawberry sugar calibrations between seasons and (2) the transferability of fat models between different meat grinding degrees. The work indicates that Raman spectroscopy has an inherent chemometric advantage for robustness. This was particularly pronounced in the strawberry case, where Raman models of around half the complexity (number of PLSR components) of NIR-based models reached similar performances and provided more interpretable regression vectors. The meat case demonstrated that the robustness of the techniques may depend on the sensitivity to different sample textures and heterogeneity. A Raman probe with a spot size of 3 mm diameter suffered from higher prediction errors than a probe with 6 mm diameter spot size. We also demonstrated an additional challenge of increasing fluorescence with longer meat storage time for Raman spectroscopy. Still, a Raman system with 6 mm spot size outperformed a reflection geometry NIR system with approximate illumination spot of 1 cm.

Few studies that compare Raman and NIR spectroscopy within food applications, focus on the differences regarding robustness. This work shows that the cost of developing and maintaining a calibration industrially should be considered to a higher extent and be part of the evaluation on final cost and implementation ease, when deciding between Raman and NIR spectroscopy or what type of sampling optic to invest in. The findings should motivate critical comparisons of the two techniques prior to long-term sensing implementation.