varroa mite detection

Why

Varroa mites are the biggest danger to colonies today

• they feed on larvae, pupae, adult honey bees

• they leave open wounds

• they consume fat body of a bee, making bees weak (immune response, energy storage, detoxification)

Suggested solution

• Datasets:
  • https://universe.roboflow.com/varroa-double/v_test
  • https://universe.roboflow.com/beproj/varroa-mites-detection--test-set
  • https://universe.roboflow.com/varroa-virus-detection/varroa-mites-detector
  • https://universe.roboflow.com/beproj/varroa-mites-detection
  • https://universe.roboflow.com/myworkspace-8q0cl/bee-detector
  • https://universe.roboflow.com/varroa-virus-detection/varroa-mites-detector
  • https://universe.roboflow.com/yolov5-object-detection-on-honey-bees-u8kya/varroua-mite-detection-in-honey-bees
  • https://universe.roboflow.com/beproj/varroa-mites-detection--train-set

App integration

• Prepare the data. For this we need to cut regions from the original frame where we detect bees - see 🐝Worker bee detection

• use AI model like yolo, get weights to detect and count mites on individual bees

• return absolute number of mites found in an image

• add http API to this model

• integrate image-splitter with this model - it manages frame photos

• store result in DB (a db migration would be needed)

• display mite infection % per frame and per hive in web-app

• generate an alert beekeeper if its above some treshold after some period of time (blocked by another issue)

Related possible detections

• detect deformed wings virus that is spread by varroa

• detect parsitic mite syndrome - uncapped larvae caused by high mite infestation

• paralysis viruses (black-shiny immobile bees)

Solution

https://github.com/Gratheon/web-app/pull/70

https://github.com/Gratheon/image-splitter/pull/17

Problem currently

• roboflow costs too much

• has too strict limits for free tier

• prediction does not work well for frame photos, even if they are sliced in parts