Analysis of Varroa Mite Colony Infestation Level Using New Open Software Based on Deep Learning Techniques.

Area of Science:

• Entomology
• Agricultural Science
• Computer Science

Background:

• Varroa destructor mites are a major threat to honey bee populations globally.
• These parasites weaken bees, reduce lifespans, and contribute to colony collapse.
• Accurate mite infestation assessment is vital for beekeeping and food security.

Purpose of the Study:

• To develop an automated deep learning system for locating and counting Varroa mites.
• To create a realistic dataset for training and validating the mite detection model.
• To improve the efficiency of Varroa mite infestation monitoring in bee colonies.

Main Methods:

• A deep learning approach using two-stage detectors with feature pyramid networks was employed.
• A novel dataset of sticky board images, including challenging artifacts and blur, was created.
• Various model architectures, hyperparameters, and image enhancement techniques were tested.

Main Results:

• The developed system achieved a mean average precision (mAP) of 0.9073 on the validation set.
• The deep learning model demonstrated high accuracy in locating and counting Varroa mites.
• The system effectively handles realistic image conditions, including artifacts and blur.

Conclusions:

• Deep learning offers a promising automated solution for Varroa mite detection.
• This technology can significantly streamline the monitoring process for beekeepers.
• Accurate and efficient mite counting supports better management of honey bee health.