The Honey.AI Project is one of the projects selected for funding through the DigiFed Innovation Pathway scheme, Low Digital Twin Application Experiments (AE). The companies involved in the Honey.AI project consist of Sonicat Systems, a Spanish industrial SME that manufactures machinery for the honey industry, and Stayia Farm, a large honey packer from Greece, who has collaborated during the project to test, validate, and extend the value chain through a future distribution agreement.

The project is framed within the vertical of agriculture and agrifood, specifically because it targets the niche market of the honey and beekeeping industry. The honey industry and the beekeeping sector have always been very traditional and conventional businesses. Honey.AI has the potential to bring a clearly disruptive innovation within this highly traditional sector, which lacks any relevant digital breakthrough to date.

Honey.AI integrates a low-cost robotised digital microscope capable of scanning autonomously a honey sample and identifying and counting the specific pollen grains by means of computer vision and deep learning techniques.

Pollen analysis in honey is used to determine the floral source, purity, and authenticity of this highly valued functional food. It consists of a very manual test that involves analysing a sample under a microscope through a 1-2 hour-long process. Pollen grains are identified and linked to the plant they came from and quantified; tasks that only highly experienced technicians can carry out. The cost of an analysis performed by a specialised laboratory ranges from 40 to 100€ and takes up to 5-6 days to get the result back.

Manual counting is expensive, time-consuming, involves human error, and the results are inevitably deferred. However, it is an essential tool for the honey industry since it is used for quality assessment, fraud screening and product labelling legal constraints. An average honey packer performs 200-500 tests per year, while a cooperative honey packer reaches up to 1,000 analyses. These figures imply thousands of euros spent every year.

Honey.AI aims to automatise the tedious work of pollen analysis for honey’s floral source authentication with higher accuracy, using artificial intelligence and robotised low-cost microscopy. This solution standardises the pollen counting measurement, reduces time, allows on-site real-time measurements, increases reproducibility/repeatability of results, and immensely reduces human dependency.

Within the project, Sonicat Systems is the company that was in charge of Honey.AI ideation and preliminary development during 2019 and 2020, thus, as an engineering company, they have been in charge of training the neural networks, and the evolved electromechanical design implementation. Moreover, they already had experience in mechanical assembling and integration since they manufacture machinery for the food industry, which implies that they have also dealt with the pre-certification analysis.

Stayia Farm is a fast-growing exporting company that operates in the honey market and other supplementary products with honey. Being among the leading producers of the finest Greek honey since 2012, the company produces and sells selected, wholesome products of monofloral, blossom honey, and honeydew honey. Stayia Farm, as a traditional honey packer, has been the low digital profile twin partner within the DigiFed Project.

During the DigiFed project, the two companies, together with Ikerlan, have enhanced the first prototype of Honey.AI, which was totally dependent on GPU cloud computing, to an embedded edge-hybrid architecture with a touchable display and a Jetson Nano, which is a series of embedded computing boards from Nvidia. Moreover, that previous version was dependant on the users’ computer and its Operative System’s constraint, e.g., having to cope with an antivirus, a firewall, a power speed, etc. This was a very relevant limitation since the device should be completely autonomous to control the positioning systems, take the pictures and upload them to the cloud. Also, the device still worked on a Raspberry Pi 4 model B board. On the scientific part, the datasets only covered the recognition of 10 specific pollen species.

As of November 2022, the new prototype is fully autonomous, does not depend on an external clients’ laptop and now includes a custom-designed Printed Circuit Board (PCB) hardware. On the software part, moving to the Edge computing part of the AI models now allows a reduction of images upload into the cloud for processing thanks to a pre-filtering step on the Edge. In addition, a new functionality for honeydew analysis has been added, with additional datasets creation and AI model implementation, as well as the number of pollen species trained has now been increased to 50. The final version has been tested and validated by Stayia Farm in Greece with their local honeys and honeydews (200 samples) and is currently under Conformité Européenne (CE) certification process. The final production costs have only increased by 400 €, which has been considered an acceptable amount.

The accuracy of both models, for Honey pollen and honeydew has been assessed by Stayia Farm and compared to the results obtained by accredited labs, with a precision above 90%. The reduction of the number of images processed in the cloud has decreased by 10 to 50% depending on the concentration and type of the honey sampled.

Honey.AI corresponds to an innovative B2B Business Model with 2 different revenues streams:

  • Device – An initial fee of 4,800€ per the automated microscope will be charged.
  • SaaS – An additional pay-per-use fee will be charged for each test conducted with 10€ per pollen analysis, and 2€ per crystals and yeasts analysis.

The support of DigiFed has been crucial during the last months, not only for funding, but also supporting the partnership with Ikerlan, as well as for the business support received and ecosystem provided. DigiFed support has been also critical to obtain a national loan from the Spanish government, as well as to raise interest from the first investors. Other funding opportunities that Sonicat Systems is analysing are ELISE, BONSAPPS and SMART4ALL, while during the project has been selected by VEDLIOT EU project, to test different edge architectures with their open source platform for IoT devices with AI applications. Partners have also been proactive in terms of dissemination through, e.g., joining Alimentaria Fair in Spain, the IoT world Congress, or APIMONDIA, the most important event of the sector, where Sonicat did a specific presentation of Honey.AI.

Sonicat Systems now is fully committed to Honey.AI commercialisation, first starting in Spain to create initial references and then expanding to other key countries such as Germany, Greece, France, Italy, Romania, Hungary, among others. Beginning the commercialisation through known markets like Spain, Greece and France will facilitate consumer acceptance and experience to demonstrate the benefits of Honey.AI to new markets/costumers. It is essential to become a reference as soon as possible.

From the Technical point of view, there are still specific milestones to reach, such as the official certification of performance by an accredited organism (Microval or AOAC), extending the datasets to include additional botanical species from all other countries, and including new functionalities, such as colour measurement. On the other hand, Honey.AI could be further upscaled to also include the geographical origin authentication service, not only the botanical source analysis. The new EU directive for honey commercialization requests to include the origin (countries) in the label, indicating where the honey comes from. Being capable of including the geographical analysis of honey by means of endemic species identification would even increase the unique value proposition of Honey.AI.