A Multimodal Machine Learning System for Non-Invasive Detection of Varroa destructor Infestations in Honey Bee Colonies

Varroa destructor infestations are a leading cause of honey bee colony collapse, yet current detection methods are invasive, labor-intensive, and unsuitable for continuous monitoring. This study presents a field-deployed multimodal sensing system for non-invasive, real-time detection of varroa infestations in honey bee colonies. The system integrates three independent sensing modalities — computer vision, acoustic analysis, and environmental sensing — on an edge computing platform costing USD 189.68, requiring no internet connection. The system was validated across nine colonies — three per infestation category — with six 30-day field deployments and three one-day deployments on clearly infested colonies, collecting 26,272 total logged observation records. A confidence-weighted fusion strategy combined predictions from all three modalities (computer vision, acoustic analysis, and environmental sensing), reducing false alarm rate from 27.0% (vision only) to 0.5% while achieving precision 0.995, recall 0.921, and F1 0.957. Field evaluation demonstrated distinct fused score distributions across healthy (median 0.0196–0.0217), mildly infested (median 0.0759–0.0889), and clearly infested (median 0.5126–0.5483) colonies, with zero threshold-positive windows across all six healthy and mild hives and 38, 37, and 34 alert episodes across the three infested hives. Temporal smoothing provided a median simulated detection lead time of 4.6 days prior to critical infestation thresholds, while field deployments demonstrated severity-consistent score separation across infestation categories. These results demonstrate the feasibility of continuous, non-invasive varroa monitoring under real-world agricultural conditions. The proposed system provides a scalable, sub-$200 approach that addresses the accessibility gap in automated varroa monitoring, where prior systems have exceeded USD 500 per hive, and supports the development of smart agricultural monitoring systems.