The quality control of food products is very important to determine their composition and nutrition facts, and to detect eventual frauds and adulterations in raw or industrialized products. For example, bovine milk can be adulterated by addition of foreign products, in order to increase the volume or to extend the expiration date, with economic and sanitary impacts. This work presents the characterization of milk using ultrasound and neural networks techniques. An ultrasonic measurement cell was used to obtain the propagation velocity, attenuation coefficient and density of milk liquid samples with different fat contents and added water. Samples were calibrated by using conventional methods employed in the dairy industry. Artificial neural networks were designed to output the fat content and added water of samples from the information of the experimental parameters measured by the cell. The algorithm resulted in more than 95% of correct classification, with resolution of 0.1% in the determination of fat content. For the determination of added water, the resolution was 1% (between 1 and 10% of added water) and 10% (between 10 and 60% of added water).
Ultrasound; Artificial Neural Networks; Milk
