Radio Frequency Tags, A.I. Designed to Detect Unsafe Food

Infant formula on store shelves (ParentingPatch, Wikimedia Commons)

15 Nov. 2018. A computer engineering team designed a simple, inexpensive system for detecting food quality with radio-frequency tags applied to consumer products and machine learning algorithms. Researchers from Massachusetts Institute of Technology describe the system in a paper given at this week’s ACM’s HotNets 2018 conference in Redmond, Washington.

Researchers from the Signal Kinetics lab, part of MIT’s Media Lab, led by computer scientist Fadel Adib are seeking more practical ways for consumers to directly assess the quality of food products on store shelves with today’s technologies. The public health issue of food safety is underscored by tainted baby formula, a problem that emerged in China in 2008, and a continuing problem of alcoholic beverages mixed with cheaper industrial methanol leading to blindness in several countries.

“In recent years,” says Adib in an MIT statement, “there have been so many hazards related to food and drinks we could have avoided if we all had tools to sense food quality and safety ourselves. “We want to democratize food quality and safety, and bring it to the hands of everyone.”

One of the everyday wireless technologies proposed for a solution is radio frequency identification, or RFID tags, printed on plastic and applied to increasing numbers of retail items. RFID tags contain identifying and other product data — e.g., size, price, color, location — in a standardized format, which respond to UHF radio signals sent from and transmitted back to a reading device. These so-called passive RFID tags require no additional power source, but respond to electromagnetic signals from the reader on demand. A survey in 2016 showed three-quarters of retailers were implementing RFID tags in their stores.

Adib and colleagues propose employing this technology with no modifications to also gauge product quality. Their idea takes advantage of interactions of the reader’s electromagnetic signals with the contents of the food in the package. These interactions distort the returned signal, and while the distortions do not affect return transmissions of data on the tag, they can be captured and analyzed separately. To capture the returned distorted signals, the researchers modified the readers to receive signals on a wider frequency range.

The team’s system, which they call RFIQ, extracts pertinent data from distortions in the returned signals and sends then to a database in the cloud containing entries of readings from similar food items with RFID tags. The researchers envision manufacturers could provide the initial data for RFIQ. Data from the returned signals would then be transmitted to the database, where they would be evaluated for indications of spoilage or adulteration. Machine learning algorithms, a form of artificial intelligence, would train the initial database, conduct the assessments, and use the submitted data from consumers to refine its analysis over time.

The researchers pilot-tested RFIQ with two types of products, alcohol and infant formula. The team mixed in fixed percentages of methanol — 0, 25, 50, 75, and 100 percent — to pure alcohol in plastic bottles. The team tested the different blends in 3 locations and 10 different bottles. Results show the system returned 97 percent accurate detection of bottles with adulterated alcohol. With infant formula, the researchers added contaminants of melamine — the formula contaminant found in China — in concentrations from 0 to 30 percent to an off-the-shelf brand. The results show 96 percent accurate identification of adulterated formula, missing only the lowest (1.25%) concentration of melamine.

While initial results are promising, say the researchers, a lot of work remains. The team highlights issues with various product types, packaging, signal range, mobility, and power of the machine-learning algorithms that need further study. For further development, the RFIQ system may also beyond safety detection to identifying the nature of the contaminants.

More from Science & Enterprise:

• Munich Airport, MIT Lab Partner on Smart City
• NVIDIA, Scripps Partner on Medical Deep Learning Applications
• Wireless Solar Power, Data Transmission Designed for IoT
• Algorithms Advance Assistive Control for Spinal Cord Injury
• Hand-Held Device Inspects Individual Retina Cells

*     *     *