--- tags: machine learning --- # Scanning bees to get more honey ### Group members: Manuel Leibetseder, Maike Rieger, Andreas Roither, Markus Seiberl ### Description There are mites that have no natural enemies. These mites attack the bees and suck out their blood until they die of anaemia. In principle, the goal of the project its to take photos of frames with bees with the help of an app and calculate how many % of the bees on the frame are infested by the varroa mite (also called varroa destructor). Beekepers can then decide to use appropriate meassures to save their bees. ![mite](https://upload.wikimedia.org/wikipedia/commons/a/a6/Varroa_Mite.jpg) ### Possible Outcomes - ~70% Accuracy, 100% is not needed since the beekeeper has to act even if there is a low amount of mites. - The time might be an important feature in ### Data Bee image datasets: https://www.kaggle.com/search?q=varroa https://www.kaggle.com/jenny18/honey-bee-annotated-images Data of honey bee annotated images: | File names | Date captures | Subspecies | Health of a bee | |--------------------|---------------------|-------------------------------------------------------------|----------------------------------------------------------| | 5172 unique values | 02.07.18 - 08.09.18 | Italian honey bee 58% Russian honey bee 10% Other (5) 32% | healthy 65% few varrao, hive beetles 11% Other (4) 23% | ### Features that might help - Health of the bee - Species - Date - Time From Moodle: **metrics to quantize** - accuracy xx% what is a good accuracy * How possible outcomes of the project could be quantified, e.g.: "X% accuracy", or "using information X outperforms using information Y in ML (X% vs Y% accuracy)". This info is important to a priori check that you try to solve "the right problem" and try to solve your problem "the right way". * Which data you plan on using for your problem: what does the data look like? Consider giving an example! What are it's features? How many data do you have? (Nr of samples, timespan they cover, etc). If you need to record data yourself, state how and when you plan on doing it. * A list of features you think might be helpful in solving your problem.