featured photo
Why is AI invaluable for food quality?

5 minutes read.

We are witnessing ever increasing standards of food safety and quality, driven by consumer demand for healthy, natural and organic food. Food producers are also pressured to constantly increase the efficiency in production, as an answer to the growing market pressure.

Vast amounts of food are wasted throughout the entire food value chain, from farming to production to logistics, through to consumers. Food and agriculture organization of United Nations estimates that 30 – 40 % of food is wasted before even reaching the market.

One of the main concerns of food producers is to avoid releasing bad quality products that could damage reputation and potentially health of consumers. Recalls of bad quality products cost the food industry an average of 10 MUSD per case, not including the financial impact of the brand damage. So, industry sets very high thresholds on what is considered safe and good quality food. Some food is wasted even though it is good for consumption, leading to much higher costs in production and putting additional pressure on food producers, who are already fighting low margins. One solution to the food waste is development of a circular economy, where wasted food is used in other production chains, as source of raw material or as animal food. This, however, doesn’t reduce waste in the value chain. Food industry is also investing in automation and more stringent quality checks, but quality control is often done manually, which is costly and error prone.

Leading players in the food industry are thus rapidly switching to machine learning and computer vision to solve some of the problems. For example, Kraft Heinz CIO developed strategy to include AI tools in all steps of business processes in his organization – from sales and marketing to supply chain and manufacturing, all with the goal to reduce costs and decrease waste.

Machine learning in food safety can help more precisely detect bad quality products, contributing to the waste reduction, and can control production equipment to automatically remove bad quality products from production lines. It can help collect and trace quality parameters through the production and pinpoint root causes of bad quality. It could also help detect products whose shelf life can be extended in retail. Finally, it can be used by consumers, to test food quality before consumption.

We in Createsi are committed to contribute to better food quality through application of advanced machine learning technologies. Here are some of our results.

In the first set of examples, algorithms take photos of different fruit as an input and sort good from bad ones, identifying at the same time image features which contributed most to the decision.

Apple 1Apple 1 resultOrange 1Orange 1 ResultOrange 2Orange 2 result

First image in each pair shown above is an original fruit image introduced at the input of the machine vision algorithm, and a second image shows an output, with the heatmap identifying area of the bad quality. It is important to note that the algorithm doesn’t segment different areas in the original image by discoloration only. It is trained to intelligently seek for the regions which indicate bad quality. We deliberately use obvious examples here to show the basic principles, but using different types of cameras (for example infra-red or ultra-violet), these algorithms are capable of detecting quality issues which are difficult for humans to detect with the naked eye. Furthermore, combining visual analysis with other information collected throughout the food production chain (for example, environmental parameters, production process measurements, etc.), it is possible to find correlations indicating causes of bad quality and leading to production process improvements to avoid quality issues whatsoever.

The second set of examples show application of similar algorithms for detection of irregular shapes. In this case, images of biscuits are presented at the input, and the algorithm is trained to detect shape irregularities and cracks.

Cookie 1Cookie 1 resultCookie 2Cookie 2 ResultCookie 3Cookie 3 result

It is clear that many other quality issues can be identified in the similar way, like discoloration, undesired change of color, lack of some specific features etc. Combination of advanced sensors/cameras and smart application of machine learning have almost unlimited potential to improve handling of food quality issues.

Createsi team is very interested to hear about your challenges related to food quality and we are passionate to help you find the best solutions. Our value is in the combination of system integration and engineering skills in manufacturing and extensive experience in application of machine learning and artificial intelligence. This experience mix allows us to provide end-to-end solutions for difficult practical problems.

Please feel free to contact us via e-mail or through social media accounts.

Stay tuned for the followup blogs, which will address other use cases and describe in more details some of the machine learning technologies mentioned here.


Lazar Kulašević
On July 18, 2019 at 10:10, Lazar Kulašević wrote:

I had fun reading this article. It is both informative and educative.

Tanishk Sharma
On July 18, 2019 at 12:26, Tanishk Sharma wrote:

Wow! This is great stuff!

On July 18, 2019 at 22:10, Nemanja wrote:

Really interesting post, can't wait to read the next one... 😀

All fields are required.

Leave a comment

All fields are required.