Research library

Automatic detection and classification of honey bee comb cells using deep learning

In a scenario of worldwide honey bee decline, assessing colony strength is becoming increasingly important for sustainable beekeeping. Temporal counts of comb cells with brood and food reserves offer researchers data for modelling colony dynamics and give beekeepers information on colony strength, an indicator of colony health and honey yield. Counting cells manually in comb images is labour intensive, tedious, and prone to error. The authors developed DeepBee©, free software capable of automatically detecting cells in comb images and classifying their contents into seven classes: eggs, larvae, capped brood, pollen, nectar, honey, and other. Using Circle Hough Transform and the semantic segmentation technique, a cell detection rate of 98.7% was achieved — 16.2% higher than the best result found in the literature. Thirteen CNN architectures were trained and evaluated for cell classification; MobileNet revealed the best compromise between training cost (~9 s per full comb image) and accuracy (F1-Score of 94.3%). All CNN models, source code, and datasets are publicly available.