Use of a NIR portable spectrometer for the assessment of fat/dry matter and protein/dry matter ratios on whole wheels of Parmigiano Reggiano cheese

INTRODUCTION
A method has been developed for rapid and non-destructive characterisation of whole cheese wheels by means of a portable NIR spectrometer; since for some parameters a good correlation was found between the composition of the rind and that of the inner part of the cheese, the inner compositional parameters of the wheels were used as reference values for the development of calibration models based on the spectra obtained from the scans on the rinds of the whole wheels.
For this purpose a Micronir 1770 JDSU spectrometer was chosen . This device works in diffuse reflectance mode and uses a linear variable filter (LVF) component mounted over a diode array detector.
EXPERIMENTAL
The instrument is distinguished by its small size, low weight (60 g) and short analysis times (a few seconds) per sample and has been engineered to estimate the cheese composition at batch level so as to make the use functional and ergonomic in storehouse environment, and to automate the flow of data from elimination of bad spectra to the composition estimation at batch level based on scans of a high number of wheels. The instrument was mounted on an ergonomic handle on which were placed the controls of the main measuring functions. A system of sound signals allowed the operator to know in real time the progress of the sampling without entering the computer that controls the apparatus. At the same time it has been developed a software that can collect and elaborate the individual production lots spectral data, without PC intervention of the operator during the individual sampling sessions. The setup of the predictive models was performed on a calibration set, consisting of about 250 samples, representative of the production of Parmigiano Reggiano at 12-18 months.
RESULTS AND DISCUSSION
The set spectral collected in the laboratory made it possible to optimize the signal processing and the development of predictive models related to moisture, fat, protein, fat/dry matter, and protein/dry matter ratios .The estimate of these last two parameters has proven particularly robust and reliable, also because these ratios are less sensitive to moisture gradient existing between the outside of the wheel, where the spectral measurements are carried out, and the inner, on which the reference measurements were carried out.
The developed prediction models showed very low, non significant bias, with a standard prediction error on the individual scans of 1.21 and 1.36 respectively for fat/dry matter and protein/dry matter ratios.
These results were considered adequate, on account of the reduction of random error of the averages, to evaluate the batch composition based on a sampling plan of 5 scans for wheel and around 70 wheels per batch.