External links
Abstract
Monitoring beehive health is a major challenge in beekeeping, including detection of Varroa destructor, hive beetles, ant problems, and missing queens. The paper proposes a deep-learning methodology for recognizing beehive health abnormalities using a modified MobileNet model. Three optimization algorithms—Adam, Nadam, and SGD—are compared on a benchmark beehive dataset, and the MobileNet model with Adam optimizer achieved reported testing accuracy of 95% for classifying the targeted abnormality classes. The study positions AI, computer vision, and IoT as components of warning, prediction, and recognition systems for hive health and bee behavior monitoring.
Relevancy to Gratheon
For Gratheon, this paper is a useful reference for lightweight vision models that classify multiple hive-health abnormality classes rather than a single pest or behavior. The MobileNet focus is relevant to edge or mobile deployment, while the abnormality categories map directly to beekeeper-facing alerts. It should be treated as a benchmark and design reference rather than a field-ready solution: Gratheon would need to verify dataset representativeness, class definitions, and real-apiary performance before using similar models in production.