Computer Vision-Based Assessment of Insect Activity at Beehive Entrances Using YOLOv8n

Abstract

Honeybee colonies are increasingly threatened by predators such as wasps, making continuous and automated hive surveillance essential for sustainable precision apiculture. Conventional manual monitoring is labor-intensive, subjective, and unsuitable for real-time deployment in large-scale apiaries. This study presents a computer vision-based framework for automated monitoring of beehive entrance activity using the YOLOv8n object detection model. The proposed system identifies four operational classes, namely worker bees, pollen-carrying bees, drones, and wasps, while incorporating an event-driven intrusion detection mechanism that automatically generates alerts for wasp presence based on configurable confidence thresholds. The framework employs a YOLO-formatted dataset comprising 8,807 training images, 2,201 validation images, and 300 testing images, and integrates data validation, model inference, visualization, alert generation, and performance analysis into a unified Streamlit-based application for rapid deployment and demonstration. Experimental evaluation demonstrates reliable detection of beehive activities with high precision, achieving precision–recall area under the curve (PR-AUC) values of 0.905 and 0.835 for worker bees and pollen-carrying bees, respectively. During testing, the system successfully detected 21 wasp intrusion events with confidence scores exceeding the predefined alert threshold, while producing annotated images, timestamped alert records, confusion matrices, and precision–recall curves to improve interpretability and operational decision-making. Owing to its lightweight architecture, rapid inference capability, and explainable output generation, the proposed framework is suitable for edge-enabled smart beekeeping applications. Future work will incorporate multi-object tracking, temporal behavior analysis, and embedded deployment to further enhance real-time hive security and colony health monitoring.