characterizing dairy products
TRICLOPS spectrometer represents a non-destructive and rapid method in the dairy industry, to accurately quantify the fat content in milk.
Transmission spectroscopy with TRICLOPS allows non-contact and rapid characterisation of many liquid samples. Moreover, its broad spectral range (400-4500 nm)provides simultaneous access to a wide range of absorption features, from colour information in the visible, to vibrational signatures in the IR.
This unprecedented range is unlocking new possibilities inmaterial identification and quantification. In this application we present a simple demonstration of how the optical absorption of milk changes with composition.
The spectra show a linear dependence on the fat content over the covered spectral range. These measurements were used to train a regression model, which was able to accurately determine the fat content of milk from subsequent measurements.
Introduction
Milk is a complex nutritional liquid mainly composed of water, fats, proteins, and lactose. Milk composition analysis is crucial for several reasons, spanning quality control, nutritional assessment, economic valuation, and regulatory compliance.
The fat content in milk is a critical parameter influencing its nutritional value, flavor, and economic value. Accurate measurement of milk fat is essential for quality control, regulatory compliance, and to ensure consistency in dairy products.
Traditional methods for analyzing milk composition are often time-consuming and require the use of chemicals. TRICLOPS Fourier transform spectrometer offers a non-destructive, rapid, and efficient alternative for milk analysis.
Figure 1
Measurements
An experimental analysis was conducted mixing different quantities of whole and skimmed milk with equal nutritional parameters, differentiated only by their fat content (3.6 and 0.1 g/100ml), in this way the differences in the samples were related only to the variation in the fat content.
In total 36 different fat concentrations were measured in transmission mode, using a cuvette of 1 mm thickness and a halogen light for illumination.
Experimental results
The spectra were obtained using TRICLOPS in full detection range (400–4500nm) with fine (10 cm-1) resolution averaging 30 scans (15 seconds total collection time), the corresponding intensity spectra for the different fat contents are shown in Figure 2a).
The light from the illuminating halogen lamp in absence of the sample was used to calculate the absorbance spectra. We then applied a pre-processing pipeline (normalization → wavelength set reduction → baseline correction) in order to reduce sample-to-sample variability due to experimental factors such as improper sample alignment.
The pre-processed spectra, in Figure 2b), show linear variations correlated to the fat concentration from 1000–4000 nm.
Figure 2: a) Raw spectra from TRICLOPS b) Pre-processed spectra
Data analysis
A predictive model was calibrated using a Partial Least Squares algorithm, validated by leave-one-out cross validation. An additional analysis on the contribution of the single wavelengths was performed to reduce the predictors set, maximizing the model performances.
The considered figures of merit are the root mean squared error of calibration and cross validation (RMSEC, RMSECV), the mean average error of cross validation (MAECV) and the coefficient of determination (R^2). The results are shown in the table and Figures 3 and 4.
Figure 3 (on the left): Fitted vs. Labeled fat concentrations
Figure 4 (on the right): Predicted vs. Labeled fat concentrations
Conclusion
The samples used for this example are overly simplistic, and not exclusively varying fat content. Nonetheless, this application note shows the potentiality of TRICLOPS in the dairy industry.
A spectral analysis for the quantification of fat content was performed, finalized in a predictive model characterized by a mean average error in the CV predicted values of 0.037 g/100ml.
Discover TRICLOPS Spectrometer
