TE-FOOD joins R&D project to develop AI based meat quality checker application

TE-FOOD was invited to a consortium of companies to take part in an R&D project with one of the main goals to develop a meat quality checker mobile app for consumers. The consortium won a national grant of over 1 million USD in Hungary.

The goal

Often people face the dilemma that a previously bought (and maybe unfreezed) meat is still recommended to consume or not.

The core aim of the project is to develop a method and tools which can be used by consumers to determine if a meat is edible by using their mobile phone.

There is a decent research literature about the meanings of colorization and surface alterations of meats. Basically this is what many people do automatically every day before starting to make a meal. They visually check the meat‘s color, surface, consistency, the smell, the information on the label, compare against their knowledge and experience, and they decide if the meat is edible. However, most people don’t have the necessary knowledge and experience.

We believe today’s technologies can help these people.

The project

During the recent years, the photo quality in average mobile phone cameras became significantly better. Meanwhile, machine learning technologies like artificial neural networks became much easier to use. These evolving technologies provide possibility to create everyday applications which were formerly available only in laboratories.

The process

• Users need to download a mobile app. They need to provide some basic information (type of meat, when they bought it, how they stored, etc.), then take a photo with the app about the meat.
• Since lights, shadows, environment and camera quality can be different, a color benchmarking card is needed to be placed beside the meat. The color benchmark card’s colors are used to standardize the color scheme of the photo. The project progress will show if different bechmark cards will be necessary for different animals (e.g. cattle, poultry, fish) or not.
• When the photo is taken, it’s sent to the neural network for analysis. The machine lerning algorithms use all former analyis results as well as professional knowledge to determine edibility according to the provided data.
• The user gets a suggestion from the system, if according the analysis the meat is edible or not.

Certainly this is only a suggestion (as there are contaminations an optical examination can not detect), and the neural network needs huge amounts of data to compare the results, but as in machine learning systems, the algorithm becomes more and more intelligent as the more people use it.

Next steps

After the alpha release of the app, we plan to reach out to the community for help teaching the application. As in machine learning a lot of examples needed to fine tune the system, we also plan to contact research facilities and quality control organisations for contribution.

TE-FOOD plans to implement the meat quality checker into its ecosystem, and embed token payment possibilities. This enables us to reward users at the start and also to charge costs in a later phase. At this moment, the exact costs/rewards levels can not be determined.

We are excited to progress with this project as soon as possible, as it gives TE-FOOD a strong B2C use case, and a possible new revenue source on the long term.

UPDATE (May 13):
First test of the benchmark cards.

TE-FOOD is the world’s largest farm-to-table food traceability solution. Started in 2016, it serves 6000+ business customers, and handles 400,000 business transactions each day.

Website: www.te-food.com

Telegram group: https://t.me/tefood
Facebook: https://www.facebook.com/tefoodintl/
Twitter: https://twitter.com/TE_FOOD