Webinar Recording: AI-driven Image Analysis for Increased Accuracy and Precision of Honey Bee Colony Health and Productivity

Accurately measuring honey bee colony health is essential for understanding how agricultural products affect pollinators. Traditional field assessments rely heavily on human estimates, which can vary widely between observers.

Dr. Ashley St. Clair, regulatory ecotoxicologist at Corteva Agriscience, shared how artificial intelligence is transforming the way researchers evaluate honey bee colonies in regulatory field studies.

Dr. St. Clair explained that regulatory agencies require researchers to monitor colony strength—including the number of adult bees and developing brood—to determine whether crop protection products have adverse effects on honey bee colonies.

While current methods ask trained observers to visually estimate how much of each honeycomb frame is covered by bees, decades of research have shown that people often underestimate bee populations and produce inconsistent results. Factors such as observer bias, fatigue, lighting conditions, and bee clustering can all reduce accuracy.

To address these challenges, Corteva researchers developed an AI-powered image analysis system that replaces visual estimates with automated analysis of high-resolution photographs. Using machine-learning models trained on hundreds of annotated hive images, the system can identify and count individual bees while also analyzing colony features, such as capped brood, developing larvae, nectar, pollen, and honey stores.

The results have been promising. The bee-counting model achieved greater than 95% accuracy and closely matched manually verified bee counts, while substantially reducing the variability seen with traditional visual estimates. In contrast, beekeeper estimates frequently underestimated bee numbers and showed much greater variation, particularly when frames contained large numbers of bees.

Beyond improving bee counts, the AI system is also being trained to evaluate overall colony organization by identifying the contents of individual honeycomb cells. This capability allows researchers to measure multiple indicators of colony health from a single image while creating a permanent digital record that can be reviewed and verified later.

Dr. St. Clair noted that image quality is critical for reliable analysis, so standardized photography protocols have been developed alongside the AI models. Although the technology is still being refined, Corteva plans to begin using the system in regulatory field studies in 2028.

By replacing subjective estimates with consistent, data-driven image analysis, AI has the potential to improve the accuracy, precision, and reproducibility of honey bee health assessments while providing regulators and researchers with more reliable information for protecting pollinators.

Learn more by watching the full AI-driven Image Analysis webinar.

Presenter