Beekeepers nowadays have to worry a lot about a species of parasitic mites called varroa. A beekeeper has to monitor the amount of varroa in each beehive and has to count how many dead ones there are.

This project aims to eliminate the need to manually count the amount of varroa for each beehive. Instead we want to use classical computer vision algorithms combined with machine learning.

In a first step the script uses thresholding and the opencv blob detector to find possible varroa parasites. This process requires the varroa to be on a white background and tries to mask out anything else. This process will return a set of small images containing the found blobs. The AI will now classify all the blobs as either a varroa hit or a false positive and return the amount of them. Alternatively you can go manually through all the possible hits to extend the dataset or evaluate the accuracy of the AI.

I made a small GUI so we can track the training progress. It provides a small selection of filters and some examples. It also allows you to save your models and load already trained ones.

you can use the train.py script which only contains:

import Source.GUI_Train as GUI

GUI.start_training_GUI(data_path=r"PATH/TO/DATA")

To actually count the varroa in a picture you can use the batchprocessing command which you have to provide with a path to a directory which contains all pictures which should be processed. This process right now will just use a already trained model in the Model directory.

import Source.Utils as utils

utils.batchProcessing(r"Path/to/Images")

This process will just output the results in the console. Will probably ad csv, json or txt files for that at some point.

I also provided a small script which allows you to easily classify a load of data in a short amount of time.

import Source.Utils as utils

img_dir = r"IMAGE PATH"
dataset_dir = r"Data"

utils.Img_to_Dataset(img_dir, dataset_dir)

This script allows you to classify possible varroa matches in a picture. It will show you a sample and you have to press the right key:

• "a" to label a sample as varroa,
• "d" to tag it as a false positive,
• any other key to skip the sample

The labeled images are stored at the location: dataset_dir

This code works on python 3.7 with:

• tensorflow 2.0 (with keras)
• numpy
• matplotlib
• PyQt5
• pyqtgraph
• opencv