Every beekeeper knows that the Varroa mite is a bee’s worst enemy. These tiny blood-suckers drain the life from their hosts, and in the process, pass viruses on to the hive. They’re a worldwide phenomenon, initially affecting Apis cerana in Asia before spreading to Apis mellifera and the rest of the apiary world, with the exception of Australia.
The question for beekeepers now isn’t, “Do my hives have mites?” but rather, “How bad is the infestation?”
Best practices recommended by the Honey Bee Health Coalition call for no less than a quarterly screening of broods for the presence of mites. The usual practice is to collect a small sample of bees–usually about 300–and wash them in alcohol or soapy water. Unfortunately, this sacrifices these tireless workers for the greater good of the hive. But by straining the dead bees from the water, the mites can be counted and a percentage of the infected calculated by dividing the number of mites by the size of the sample. Another common method of counting the mites is a controlled shake in powdered sugar, dislodging the mites, and allowing later counting by dissolving the sugar in water. The advantage of this technique, of course, is that the sugar is harmless to the bees.
Manual sampling problems
In both cases, the sampling must be performed carefully to ensure uniformity. If 350 bees are collected the first time, 315 bees the second, and 280 bees the third, and 250 the fourth, a count of 5 mites each time would mistakenly lead a keeper to estimate that the Varroa population was in decline. Like a scientist in a lab, the more careful the sample and count, the more accurate the resulting estimate–though of course, truly precise counts aren’t possible in the field.
This is a major problem. As brood numbers increase toward their peak population, Varroa destructor tracks this curvilinear function with a slight delay. And when just 3-5% of the bees are stricken by mites, the hive is already in serious trouble. Given that Varroa’s presence in your hives is a given, that doesn’t provide much room for error.
Unfortunately, concerns about long-term efficacy and insecticide synergism prevent keepers from liberally applying miticides. Instead, these controlled dosages must be administered only as needed to control the population of Varroa, reinforcing the need for careful frequency analysis.
Is machine learning the future?
Hightech beekeeping is the future, and systems like Bee Smart that provide accurate, real-time information to beekeepers are vital to maintaining healthy hives. As trendwatchers and futurists, we look at innovations, making imaginative leaps forward to predict the trend of development over the next decade or so.
Bee Smart is one such trend, and we think that recent advances in machine learning and image recognition may soon offer a solution that works synergistically with big data-driven apiaries. By collecting thousands of pictures of bees in hives around the world, a Swedish entrepreneur, Björn Lagerman, wants to teach an artificially intelligent algorithm to recognise these killers and help beekeepers protect their hard-working friends.
Imagine machine learning as a vertical structure of processes and algorithms. At lower levels, basic data is captured and assessed. In this case, a picture of a brood frame would be broken down by the algorithms to identify individual bees. It might first look for things it can identify as a thorax here and an abdomen there, a set of wings or a head. It’ll judge this by looking for simple edges and shapes. In just fractions of a second, it will compare this image with hundreds of the thousands of others to which it’s been exposed and by dint of which it has learned to distinguish bee from the background.
This first layer thinks it has identified 47 individual bees. The second would work with this assumption, identifying each bee’s orientation and size. It will confirm this count and begin working on identifying tell-tale signs of Varroa: stunted wings and decapped broods. It’s ‘thinking’ about categories like ‘sick’ and ‘healthy’ here, and again, in just fractions of a second, it’ll complete its work, passing this information to the next higher level.
At each stage, this deep learning process imitates the function of the human brain, abstracting, assessing, assuming, and analysing. In no time, it’s counted mites on the bodies of bees, sorted healthy and sick, and assessed the need for intervention.
Big data and machine learning: the new beekeeping
This is the system Lagerman’s pushing for, hiring a team previously tasked with having artificial intelligence and machine learning auto-identify corporate logos in photographs. Lagerman’s also brought together an impressive group of advisors: Randy Oliver, of scientificbeekeping.com, Atsuto Maki and Serge Belongie, professors with expertise in deep learning and computer vision, and Joachim de Miranda, a professor of entomology who specialises in the honey bee.
This is real tech–doable science.
The first steps, of course, are to provide the machine mind enough photos to train it. Right now, Lagerman and his team are building the technical capacity to analyse the photos. Soon, they’ll soliciting brood pics from beekeepers worldwide. The more input, the better the output–and the machine mind can keep learning, refining its performance as it works.
This could be a major breakthrough for beekeepers, but we think we see a brighter future than he imagines.
His vision is to provide a mobile-based app that allows a keeper to take a simple pic of his or her bees and run this analysis. By itself, that’s simply awesome! Lagerman’s goal, of course, is to help keepers keep on top of their worst enemy, and ultimately, to identify bees that are resistant to Varroa destructor so that we can stop using miticides altogether. This is the only long-term solution that’s viable for the bees.
But imagine combining Lagerman’s machine learning with something like Bee Smart. A tiny camera and flash could take photos in real-time, assess mite numbers accurately, and provide a simple infographic on your mobile or PC. Every day, you’d know mite numbers for each hive, receive actionable insights about integrated pest management (IPM), and know how many Varroa resistant workers were present. Each hive, then, would provide data about hive temperature and humidity, brood temperature, and hive acoustics, as well as accurate frequency analysis of the mite population and treatment recommendations.
That could revolutionize the fight against these deadly pests.
