Determination of the lactose content in low-lactose milk using Fourier-transform infrared spectroscopy (FTIR) and convolutional neural network.

Demand for low lactose milk and milk products has been increasing worldwide due to the high number of people with lactose intolerance. These low lactose dairy foods require fast, low-cost and efficient methods for sugar quantification. However, available methods do not meet all these requirements. In this work, we propose the association of FTIR (Fourier Transform Infrared) spectroscopy with artificial intelligence to identify and quantify residual lactose and other sugars in milk. Convolutional neural networks (CNN) were built from the infrared spectra without preprocessing the data using hyperparameter adjustment and saliency map. For the quantitative prediction of the sugars in milk, a regression model was proposed, while for the qualitative assessment, a classification model was used. Raw, pasteurized and ultra-high temperature (UHT) milk was added with lactose, glucose, and galactose in six concentrations (0.1-7.0 mg mL-1) and, in total, 432 samples were submitted to convolutional neural network. Accuracy, precision, sensitivity, specificity, root mean square error, mean square error, mean absolute error, and coefficient of determination (R2) were used as evaluation parameters. The algorithms indicated a predictive capacity (accuracy) above 95% for classification, and R2 of 81%, 86%, and 92% for respectively, lactose, glucose, and galactose quantification. Our results showed that the association of FTIR spectra with artificial intelligence tools, such as CNN, is an efficient, quick, and low-cost methodology for quantifying lactose and other sugars in milk.

A machine learning proposal method to detect milk tainted with cheese whey.

Cheese whey addition to milk is a type of fraud with high prevalence and severe economic effects, resulting in low yield for dairy products, nutritional reduction of milk and milk-derived products, and even some safety concerns. Nevertheless, methods to detect fraudulent addition of cheese whey to milk are expensive and time consuming, and are thus ineffective as screening methods. The Fourier-transform infrared (FTIR) spectroscopy technique is a promising alternative to identify this type of fraud because a large number of data are generated, and useful information might be extracted to be used by machine learning models. The objective of this work was to evaluate the use of FTIR with machine learning methods, such as classification tree and multilayer perceptron neural networks to detect the addition of cheese whey to milk. A total of 520 samples of raw milk were added with cheese whey in concentrations of 1, 2, 5, 10, 15, 20, 25, and 30%; and 65 samples were used as control. The samples were stored at 7, 20, and 30°C for 0, 24, 48, 72, and 168 h, and analyzed using FTIR equipment. Complementary results of 520 samples of authentic raw milk were used. Selected components (fat, protein, casein, lactose, total solids, and solids nonfat) and freezing point (°C) were predicted using FTIR and then used as input features for the machine learning algorithms. Performance metrics included accuracy as high as 96.2% for CART (classification and regression trees) and 97.8% for multilayer perceptron neural networks, with precision, sensitivity, and specificity above 95% for both methods. The use of milk composition and freezing point predicted using FTIR, associated with machine learning techniques, was highly efficient to differentiate authentic milk from samples added with cheese whey. The results indicate that this is a potential method to be used as a high-performance screening process to detected milk adulterated with cheese whey in milk quality laboratories.

On the utilization of deep and ensemble learning to detect milk adulteration.

BACKGROUND: Fraudulent milk adulteration is a dangerous practice in the dairy industry that is harmful to consumers since milk is one of the most consumed food products. Milk quality can be assessed by Fourier Transformed Infrared Spectroscopy (FTIR), a simple and fast method for obtaining its compositional information. The spectral data produced by this technique can be explored using machine learning methods, such as neural networks and decision trees, in order to create models that represent the characteristics of pure and adulterated milk samples. RESULTS: Thousands of milk samples were collected, some of them were manually adulterated with five different substances and subjected to infrared spectroscopy. This technique produced spectral data from the milk samples composition, which were used for training different machine learning algorithms, such as deep and ensemble decision tree learners. The proposed method is used to predict the presence of adulterants in a binary classification problem and also the specific assessment of which of five adulterants was found through multiclass classification. In deep learning, we propose a Convolutional Neural Network architecture that needs no preprocessing on spectral data. Classifiers evaluated show promising results, with classification accuracies up to 98.76%, outperforming commonly used classical learning methods. CONCLUSIONS: The proposed methodology uses machine learning techniques on milk spectral data. It is able to predict common adulterations that occur in the dairy industry. Both deep and ensemble tree learners were evaluated considering binary and multiclass classifications and the results were compared. The proposed neural network architecture is able to outperform the composition recognition made by the FTIR equipment and by commonly used methods in the dairy industry.