Beekeepers are increasingly being advised to monitor varroa mite counts in their hives, with the goal of improved management of the external parasite. Now, an invention out of Sweden hopes to make the task of counting mites and interpreting the data even easier – by using artificial intelligence (AI) through an app on smart phones. According to its creator, BeeScanning has several advantages over other commonly used monitoring techniques and it could benefit beekeepers in many ways, beyond just counting mites.
AI, that is the use of computers to complete tasks usually requiring human perception, is now common in many walks of modern life and Swedish beekeeper Björn Lagerman has spent the last five years putting it to use in his hives.
BeeScanning is a smart phone app which requires the beekeeper to capture photos which include 3000-6000 bees per hive tested, usually requiring 12-24 photos over 2-4 brood frames. Those photos are then sent, through the BeeScanning app, to its Cloud based online AI analysis tool to count both bees and visible mites. The app then completes a calculation to provide a predicted varroa infestation rate for the hive, with the result returned within minutes, when connected to suitable internet.
The calculation is based off “extensive research” to determine how the number of mites identified through AI scanning compares to the counts which can be gained through the alcohol wash method of sampling a selection of bees.
By far the most common query from beekeepers, when they learn of the concept is, how can BeeScanning count the number of mites hidden on the underside of the bees?
“Of course we can’t,” admits Lagerman, “but we can calculate it, since we have compared the AI with the alcohol washing method and correlated it.”
“When we started this process, we compared what mite count you can find visually on an image, compared to what you can find in an alcohol wash test. We have done extensive research comparing how many mites can be found with an alcohol wash and how many with artificial intelligence. We multiply the AI count with a factor that corresponds to the alcohol wash. There is surprisingly many varroa that you can’t see, hidden under the bees.”
Currently they use a factor of about eight, but the number is always being adjusted as the AI technology develops and more research comes to hand.
How it Began
Lagerman, a high school science teacher for the past 20 years, has kept honey bees in Sweden since 1972, with about 100 colonies producing three to six tonnes of honey a season. In 2016 he began developing BeeScanning, backed by crowdfunding through a Kickstarter campaign and working alongside Örebro University’s AI department in Sweden.
Over the past five years the technology has developed considerably and the BeeScanning app has been downloaded more than 15,000 times and has 3000 active users. Most are from the United States, followed by Sweden, Germany, Ukraine, Turkey and England. Around 70 people have downloaded in New Zealand, but it is early days for the company in this part of the world.
“It is spreading organically, although in some of those markets we have done small Facebook advertising campaigns to make people aware of it. I expect it will spread more and more,” Lagerman says.
When his team of beekeepers first started supplying the university with images to detect varroa on, there was a steep learning curve for the AI. Now though, Lagerman says the AI tool is “really good” at identifying mites and picks up about 80 percent of those visible to the naked eye. As usage of the app increases, and they receive more images from around the world, the founder expects BeeScanning to only get more accurate.
More Uses
It is not just varroa that the AI scanner can identify though and Lagerman hopes that BeeScanning has a more diverse set of uses going forward.
“We have trained the app to detect 15 different objects, such as deformed wing virus, American foulbrood, chalkbrood, sacbrood and even chewed cappings. It could be a useful tool for bee health authorities to see if there is a disease outbreak anywhere, and intensify their sampling in those areas.”
Further to that, he hopes BeeScanning can improve biosecurity efforts by aiding early identification of invasive pests into a country, such as tropilaelaps mites, another parasite not yet found in New Zealand.
“If we start to look for tropilaelaps with BeeScanning, we will detect it long before anyone else will. Just by beekeepers taking images to find varroa, we will find that beast too. It could be very useful for defending borders.”
BeeScanning also offers the advantage of storing mite monitoring data in one base once calculations are completed, as opposed to the considerable manual entry required with other mite counting methods. This can be of considerable value to large beekeeping operations Lagerman believes.
“That is a unique feature of the app, in that it provides detail of what was done, where, and the result.”
Assessing and displaying information is made easier too, with BeeScanning assisting in converting hive data into more understandable forms, such as heat maps. All this can be achieved while taking into account privacy concerns of beekeepers, Lagerman says.
Give it a Go
While he understands beekeepers might be sceptical over the accuracy of the results provided by the AI technology, or the practicalities of photographing bees, Bee Scanner’s founder says the calculations used are thorough. Because New Zealand’s European honey bee varieties are similar to those encountered in the countries where the technology has been developed, there should be less refining of the AI required than in countries that have bees of more varying strains.
“We have trained the app to find varroa on our bees and also on the material that comes in. We have had 150,000 images come in from around the world, mostly of European honey bees. The more the database grows, the more reliable it is in the different circumstances.”
Lagerman says they have done “many, many, many trials” to confirm the calculations they are using are as accurate as can be, but concedes that BeeScanning does slightly undercount compared to an alcohol wash.
“It is a little less accurate than the alcohol wash, but it is good enough to give you the idea of if the colony is safe, or if it needs more attention.”
Downloaders of the app can use it free of charge on two hives and after that a subscription rate of NZD$7.99 a month, or $62.99 a year applies.
With many New Zealand beekeepers living or operating in rural areas of reduced internet speed, uploading numerous image files could prove impractical and limiting. However, the app’s performance is continually being refined as its creators seek to improve not just accuracy of mite counting, but also app performance, Lagerman says. He hopes Kiwi beekeepers will join many others around the world and test BeeScanning for themselves.
“If people are wondering if the app is reliable or not, the best thing I can do is recommend them trying it out to see if it is useful,” the creator says.
More information on BeeScanning can be found via www.beescanning.com
コメント