Data and Model Driven Beekeeping

Bees make a significant contribution to biodiversity. Pollination is indeed an essential step in the life cycle of plants, and bees contribute fully to it. However, the threats that weigh today on biodiversity herald the disappearance of pollinating insects, and in particular that of honey bees (such as theApis mellifera residing in Europe). This disappearance would be catastrophic for humans: it is estimated that nearly 35% of our food depends on the pollination services provided by bees!

Colony Collapse Syndrome colony collapse disorder or CCD, that is to say the systematic abandonment of hives) is a phenomenon that has become recurrent in colonies of honey bees. A significant decrease in bee colonies is thus reported throughout the world; she could cause up 90% loss of hives. Concerning honey bees, the causes can be multiple: toxicological, parasitic, viral, even predatory, with the appearance in recent years of the Asian hornet (Vespa velutina) then from the eastern hornet (Vespa orientalis) in the hexagon.

By raising bees, beekeepers play a key role in saving the species. To help them, they now have less invasive solutions for monitoring and predicting the state of health of hives. In particular, this article focuses on two of our research works, which use Internet of Things technologies. Internet of things or IoT) associated with artificial intelligence, computer models and simulations, in order to help beekeepers in their business practices.

Behavior of bees at the entrance to the hive

The first key to monitoring and predicting hive health is modeling bee behavior; it makes it possible to identify the strengths or weaknesses of the hive, caused by diseases, famines or predators. We can thus understand what is happening inside the hive by observing what is happening outside. For this, the image and the video are rich sources of information to be exploited, non-destructive, and which constitute a strong scientific and ecological challenge. Nevertheless, trajectory modeling had not yet been developed, which motivated our work.

Our first works focused on the detection of bees, as well as the modeling of their trajectories. The idea was to individually capture the flight paths of the bees, then to characterize the rhythm of the overall activity in front of the hive in order to deduce observations which will be consolidated and made available to scientists to be cross-referenced with beekeeping data. (famine, predator, drought…).

In practice, it involved capturing the movement of the bees with a fixed, non-invasive camera, filming the entrances and exits of the insects. Using a high resolution sensor and high acquisition frequency, image processing techniques make it possible to isolate the bee from the background.

Figure 1: Center of the body (green dot) and displacements (red, blue, purple and yellow lines) of two bees, calculated from several images.
Baptiste Magnier, Gregory Zacharewicz, Provided by the author

Next, bee counters are extracted from each frame of the video, allowing the center of each bee to be detected (the green dots on the bees in Figure 1). Then, the orientation of each bee is represented by an ellipse (related to the shape of a bee). The orientation and the size of the ellipses in the image make it possible to calculate the movements of the bees between the different images of a video (blue and red lines for bee 1 and yellow and purple lines for bee 2).

Curve of trajectories of two intersecting bees. In the image on the left, the crossing is consistent. In the image on the right, one of the trajectories stops while the other goes back to make a fork
Figure 2: Plot of inconsistent (left) and consistent (right) trajectories.
Baptiste Magnier, Gregory Zacharewicz, Provided by the author

Indeed, it is easier to follow in the videos these objects of constant sizes and same orientations rather than deformable objects. This technique also avoids confusing bees and calculating aberrant trajectories, as illustrated in Figure 2.

Different traces are thus recorded. Figure 3 is a result on a video containing 1755 frames. It shows trajectories of bees entering the hive (green lines), leaving (red), or simply passing in front (blue). Misidentified trajectories are also shown in blue. From these data, it will then be possible to study and classify the behavior of bees.

Dozens of intersecting colored lines, with no obvious pattern
Figure 3: Monitoring of bee trajectories in front of the hive, based on 1,755 images. Green lines: bees entering the hive; red lines: bees leaving the hive; blue lines: bees passing in front of the hive and poorly identified trajectories.
Baptiste Magnier, Gregory Zacharewicz, Provided by the author

In the future, bee behavior can be further interpreted by supplementing the study with machine data learning and a semi-supervised AI method.

Physical characteristics of the hive

The second decision support key for the beekeeper is the analysis of the internal health of the hive. Here the work of our team uses connected scales combined with data from several sensors (such as relative humidity and internal temperature) to analyze the evolution of the weight of each hive, as well as videos of the activity on the flight board (as presented below above).

The BeePMN project – led by our team in partnership with USEK in Lebanon, ConnectHive in France and l’Atelier du miel in Lebanon – contributes to the physical characterization of the state of the hive using apiaries’ data from databases of shared data. We have proposed a methodology based on the recognition of characteristic patterns in the weight data recorded by a hive scale. A reason can be an increase in weight, then a plateau followed by a drop in weight which would correspond to a departure of bees (by swarming for example).

Subsequently, the data collected was evaluated and processed using algorithms, which made it possible to discover recurring patterns associated with events occurring in the hive.

Connected beekeeping

Coupled with other data, these two examples could eventually be integrated into a generalized hive monitoring system. Computer models (shown in Figure 4) are automatically triggered by a series of pre-defined alerts, prompting the beekeeper to take action. For example, the information for the beekeeper of the need to feed the bees can be triggered by the passage of the weight below a reference value during a particular period in autumn or winter.

Concretely, as beekeeping tasks cannot be automated and human intervention is mandatory, the proposed system will simply help and guide the beekeeper to plan and perform more precisely several relevant tasks: breeding and reproduction of new colonies, feeding colonies, adding supers (upper part of the hive to collect honey), planning sanitation operations, controlling pests (such as varroa), planning operations such as hibernation, etc. The beekeeper will thus be able to easily monitor his colony, carry out his routine tasks, respond to alerts in the event of possible malfunctions and forecast his future supply needs.

Decision diagram to estimate whether to feed the hive or not based on its weight and the environment, and sending an alert to the beekeeper via his smartphone if the answer is positive
Figure 4: Hives, sensors, AI and smartphones/tablets form a generalized monitoring system.
Gregory Zacharewicz, Baptiste Magnier, Provided by the author

These models are based on business rules built with the help of domain experts. It is also envisaged that these business rules may subsequently evolve thanks to the contributions of the community of beekeepers using the system.

Finally, all of the above will be orchestrated and presented within a user-friendly interface on smartphone or tablet, based on the principles of gamification.

This contribution will improve the experience of amateur and professional beekeepers, reducing the risks of operating an apiary and will open the door to other available inputs (detailed weather phenomena, flowering maps, humidity, behavior of bee colonies , etc.) to extend the capabilities of the simulations. Finally, the contribution of the latest generations of digital techniques, and in particular the simulationshould pave the way for more precise beekeeping, in order to minimize invasive and synthetic treatments.

We want to say thanks to the author of this write-up for this outstanding content

Data and Model Driven Beekeeping


Check out our social media profiles as well as the other related pageshttps://www.ai-magazine.com/related-pages/