Automated classification of food products using 2D low-field NMR.

In this work, low-field proton (H) and sodium (Na) relaxation and diffusion measurements are used to detect and classify different types of food products. A compact and low-cost system based on a small 0.5 T permanent magnet has been developed to autonomously authenticate such products. The system uses a simple but efficient double-tuned matching network suitable for H/Na NMR. Various machine learning algorithms are used to classify food samples based on T-T and D-T data generated by the system, and the accuracy and prediction speed of these algorithms are studied in detail. The influence of temperature drift upon prediction accuracy is also studied. Experimental results demonstrate reliable classification of cooking oils, milk, and soy sauces.