Development of a Smart Beehive System with Integrated In-Hive Environmental Control and Deep-Learning-based Image Recognition

A smart beehive system integrating environmental control and image recognition was developed for real-time and automated monitoring of bee colony activity. The system continuously measured temperature, humidity, CO₂ concentration, and feed weight, and automatically activated a heater (winter mode) or fan(summer mode) according to preset thermal thresholds. An RGB camera installed above the hive entrance captured bee traffic, and a MobileNetV2-based deep-learning algorithm quantified incoming and outgoing bees. Two detection algorithms were compared: Algorithm 1 (three-zone model) and Algorithm 2 (four-zone model with a boundary zone for hovering correction). Experimental validation was conducted in greenhouse environments during summer and winter. The environmental control algorithm maintained internal temperature within ±2-3°C, demonstrating stable heater and fan operation. Fan activation significantly reduced CO2 concentration (1,281 ppm → 758 ppm, p<0.01), indicating effective ventilation. Algorithm 2 achieved higher detection accuracy than Algorithm 1 for in-, out-, and total-counts(71.2±36.2%, 75.0±27.0%, and 73.1±31.5%; p<0.001). Detection accuracy was greater for Bombus terrestris than for Apis mellifera due to differences in body size and hive-entrance structure. The developed system automatically stabilized the hive microenvironment while quantifying bee activity with high accuracy. These findings provide a technological basis for integrating environmental regulation and behavioral monitoring in managed pollinator systems and contribute to the development of precision apiculture and sustainable pollination management.