Artificial intelligence has taken an interesting turn at Nanyang Technological University, Singapore (NTU Singapore) where scientists have created an artificial olfactory system that simulates the mammalian nose to determine the freshness of meat.
The “electronic nose” (e-nose) uses an intricate barcode system to predict meat freshness. The e-nose uses a barcode that changes colors in response to the gases produced by meat as it decays and a barcode reader smartphone app powered by artificial intelligence.
The e-nose returned a 98.5% accuracy reading when tested on commercially packaged chicken, fish, and beef samples that were left to age. The deep convolutional neural network AI algorithm powers the device, which was compared to a commonly used algorithm being used now which only returned a 61.7% accuracy with the same samples.
The e-nose device and study findings were published in the scientific journal Advanced Materials in October 2020, showing positing that food wastage could be reduced by confirming to customers whether meat was fit for consumption. The AI and app in the hands of a consumer could produce more reliable results than simply reading the “USE BY” label found on foods in the market. The NTU scientists teamed up with scientists from Jiangnan University, China and Monash University, Australia.
The leading scientists from the project predict that the e-nose technology they have invented can be built into meat packaging for quick and simple monitoring by grocery stores, butchers and consumers. Less meat will be thrown away since the actual fit for consumption level will be produced scientifically, not guessing by dates. The result has been a patent filed for the real-time monitoring of food freshness, and now the team has partnered with a Singapore agribusiness company to extend the concept to other types of perishables.
The colored barcode that reacts to the gases released by the meat and the corresponding smartphone app barcode reader produces results in about 30 seconds. When gases produced by the meat bind to receptors in a mammalian nose, signals are generated and communicated to the brain. The brain will then collect the responses and puts them into identifiable patterns to tell the mammal whether the meat is rotten or fresh.
For the e-nose, 20 bars in the barcode act as receptors, each bar made of chitosan (a natural sugar) that is embedded on a cellulose derivative and then filled with different types of dye. The dyes will then react with the various gases emitted by the meat and change color in response, which results in a unique combination of colors that act as a scent fingerprint of sorts. The fingerprint indicates the freshness level of the meat.
For example, the first barcode contains a yellow dye that is weakly acidic. The yellow dye will change to a blue-ish color when it is exposed to nitrogen-containing compounds that are emitted by decaying meat called bioamines. The color intensity is altered with the increasing amounts of bioamines as the meat decays more and more.
Scientists in this study utilized an already-existing meat freshness classification system (fresh, less fresh, spoiled). This system is built upon the extracting and measuring of the amount of ammonia and two other bioamines found in fish packages wrapped in transparent PVC (polyvinyl chloride) packaging film and stored at 39 degrees Fahrenheit over five day intervals.
The barcodes which were built into the packaging were read throughout the process.
The AI algorithm used in this study, known as deep convolutional neural networks, was then configured to the differing images of barcodes in order to identify patterns in the scent fingerprints that indicated the levels of freshness.
In order to test the accuracy of the algorithm, scientists monitored the freshness of packaged chicken, fish, and beef by gluing the barcodes inside the packaging and storing the meat at 77 degrees Fahrenheit. The team studies over 4,000 images of the barcodes from six meat packages that were taken over time intervals of 48 hours without opening the packages.
The results yielded at 98.5 percent accuracy, which consisted of 100 percent accuracy in identifying spoiled meat and a 96-99 % accuracy for fresh and less fresh meats.