Automated Hive Image Analysis: A Comparative study on Machine Learning Approach for Bee Detection and Classification
Abstract
Honeybee populations are essential for pollination and agricultural productivity but face increasing threats from parasites, pesticides, and climate change. To address the need for effective hive monitoring, this study introduces a deep learning-based system for automated bee detection and classification. Utilizing the YOLO (You Only Look Once) architecture, the proposed method performs real-time identification of various bee types, including worker bees, pollen carriers, hornets, and queen bees. The system consists of four stages: object detection, bee isolation, feature extraction, and classification. A dataset of 5,025 annotated hive images was used to train and compare multiple YOLO versions. Among them, YOLOv8 achieved the highest accuracy—98.87% at 100 training epochs. Evaluation metrics such as confusion matrices and performance curves confirm the system's effectiveness in distinguishing between bee categories. The proposed approach offers a scalable, accurate, and efficient solution for intelligent beekeeping. Future work will focus on real-time behavior analysis, handling environmental variability, and deployment on low-power edge devices.