RUN-EU Short Advanced Programme

bee-safe-cv

A mobile app that detects Varroa mites and bee gender from photos, running the model on the phone itself. I built it with an international student team as part of a RUN-EU Short Advanced Programme on AI for beekeeping.

Python
YOLOv8
ONNX
TensorFlow Lite
Flutter
Roboflow

The problem

Varroa mites are one of the biggest threats to honeybee colonies, and beekeepers usually spot them by eye. We wanted something a beekeeper could point a phone at in the field, without needing a signal or a cloud service.

How it works

Data. We labelled bee images in Roboflow for mites and bee gender.
Training. We trained a YOLOv8 object detection model on the labelled set.
Conversion. We exported the model to ONNX, then converted it to TensorFlow Lite so it was small enough for a phone.
App. The Flutter app runs the TFLite model on the device and shows detections over the camera image.

Grid of bee images with blue boxes around each bee and smaller boxes labelled varroa around the mites — Detections from the trained model, with confidence scores for each bee and mite.

How we worked

The team ran Scrum with Trello, used Git on Bitbucket, and held regular stand-ups and demos. We presented the final app to faculty and industry partners at Howest University of Applied Sciences in Kortrijk, Belgium.

Charlie presenting the team's roadmap slide to the audience at Howest — Presenting our roadmap at Howest.