Beehive-entrance imaging and deep learning for real-time monitoring of Varroa destructor in apiculture.

Varroa destructor is a major ectoparasite threatening apiculture worldwide, especially in Taiwan. We retrofit conventional hives with an entrance imaging chamber and deploy a YOLOv5s-based detector for real-time mite detection on Apis mellifera. Trained on 1,600 annotated images supplemented with mite close-ups, the model achieved a mean average precision (mAP@0.5) of 97.4 %. Video tests at hive entrances further confirmed robust performance under motion and illumination variability. We retrofitted conventional hives with an entrance imaging chamber and implemented a YOLOv5s-based detector for real-time mite detection. To facilitate adoption, we present a per-hive bill of materials and a five-year annualized cost model, demonstrating a low annual per-hive cost suitable for apiary-scale deployment. The proposed system reduces labor-intensive inspections and enables early mite detection, contributing to sustainable and data-driven beekeeping practices.