Microfy.AI’s AI-powered microscopes are taking the pain out of quality testing for honey producers, and could soon be revolutionising fermentation process control in the beer, wine, and mycoprotein sectors.
Microfy.AI’s intelligent microscope technology is an example of how AI can enable the development of solutions to industry problems – in this instance case by automating the arduous task of counting and characterising pollen grains in honey.
The Barcelona-based deep-tech firm was originally founded in 2014 by a team of electronic engineers to exploit a patent that had been granted for the development of food processing machinery based on ultrasound technology.
One of its first projects was using its technology to liquify honey without the use of heat. During this project the engineers became aware of the challenges honey producers face around pollen characterisation and analysis, which sent the startup on a different trajectory.
“When we started testing our ultrasound technology with early adopters, we could see they were very worried about whether it would have an effect on the pollen grains in their honey. We asked them why they were so worried, and they said that it was because if the pollen grains were damaged, they would not pass the standard for classifying their honey as monofloral,” Iratxe Perales, CEO of Microfy Systems, told this publication.
Delving deeper into this subject, Microfy.AI discovered that existing pollen analysis methods were less than ideal.
“The producers explained how difficult it is to analyse pollen because it has to be done manually by a trained palynologist with a microscope. They have to count hundreds of pollen grains in the sediment and they have to identify which species they are,” she said.
The largest honey producers might have a palynologist on-site, but smaller companies cannot justify the cost. Instead they have to send samples to external labs for testing, with the results usually received within two to three weeks, said Perales.
This prompted Microfy.AI to explore the development of an AI-enabled microscope to automate and accelerate the analysis process.
“We came up with the idea of training an artificial intelligence module to act as a human and be able to recognise and count the pollen species in samples,” said Perales.
In order to be a viable solution for the industry, Microfy.AI knew the microscope had to be both affordable and easy to use, which introduced several challenges, according to Perales.
“It is difficult when working in deeptech to make solutions affordable, because usually, with cheaper components there is a compromise in performance, whether that is image quality or processing speed, so we were constantly having to contend with these technical limitations.”
However, Microfy.AI worked through these challenges, and, last year, after three years of development, was ready to introduce Honey.AI to the marketplace.
Honey.AI integrates various elements – digital microscope, a web or desktop visualisation APP, an Artificial Intelligence pipeline hosted in the cloud, and computer vision algorithms – to deliver a fully automated solution.
The intelligent microscope is able to autonomously scan honey samples and provide the user with almost immediate insights into parameters such as colour, pollen spectrum, crystallisation degree, and honeydew elements. It does this by scanning a sample on a glass slide, taking an image of the sample and sending it to the cloud, where it is passed through deep learning algorithms developed and trained by Microfy.AI. These algorithms recognise the pollen grains and classify them into more than 100 species.
“Every week we create new data sets and retrain the models to be more accurate,” said Perales.
In parallel to rolling out its Honey.AI solution in the honey industry, Microfy.AI is developing two further applications for its intelligent microscope technology: Ferment.AI and Fungi.AI.
Ferment.AI is essentially a process monitoring tool for beer and wine producers. It determines which micro-organisms are in the media; the concentration at which they are present; and whether they are dead or alive.
As with pollen analysis, this type of analysis is currently being performed manually. Perales explained: “Companies take a sample, put it under the microscope and manually count the spores or the yeast – there could be 200 or more. The problem is that this is tough, it is tiring, it is prone to human error, and it is a non-value-adding task. With our system, it takes less than five minutes and is more accurate.”
For beer and wine producers, this type of analysis is for quality purposes rather than to satisfy any industry standards. If producers do not monitor their media correctly, it can affect qualities like sparkle or flavour.
The third application, Fungi.AI, is more of an R&D tool, aimed at eliminating manual monitoring during mycoprotein development.
“At the R&D stage, mycoprotein producers have to do a lot of manual microscope work to track how the mycoprotein is growing. We have automated this so it is faster and easier,” noted Perales.
“Ultimately, that is the crux of what we do. We automate processes that already exist in the food industry and that are either done manually or not at all because of a lack of knowledge or experienced personnel. We provide the user with a plug-and-play microscope that is autonomous, affordable, already trained and easy to use.”
