Computer Vision Approaches for Automated Bee Counting Application

TL;DR

This work addresses automated counting of bees at hive entrances using computer vision and compares three strategies: a conventional motion-analysis method, a CNN-based classifier, and a YOLOv8-based object detector. Evaluations on two bee-traffic datasets (BUT1 and BUT2) show that the ResNet-50 classifier delivers the best performance, achieving 87% accuracy on BUT1 and 93% on BUT2, with 93.22% overall when combining test sets; conventional methods undercount due to cluster movements, and YOLO detectors underperform relative to the CNN approach. The study demonstrates that a lightweight CNN classifier can provide robust bee-counting suitable for edge deployment, while detector-based approaches face challenges in this context due to clustering and computational demands. The results underpin practical applications for monitoring colony health, bloom dynamics, and pesticide effects, enabling real-time trend analysis on embedded devices at around 5 Hz.

Abstract

Many application from the bee colony health state monitoring could be efficiently solved using a computer vision techniques. One of such challenges is an efficient way for counting the number of incoming and outcoming bees, which could be used to further analyse many trends, such as the bee colony health state, blooming periods, or for investigating the effects of agricultural spraying. In this paper, we compare three methods for the automated bee counting over two own datasets. The best performing method is based on the ResNet-50 convolutional neural network classifier, which achieved accuracy of 87% over the BUT1 dataset and the accuracy of 93% over the BUT2 dataset.

Figures (3)

• Figure 1: Illustration of the measurement setup.
• Figure 2: An illustration of the samples from the BUT1 and BUT2 datasets.
• Figure 3: Confusion matrix of the selected ResNet model trained over all classes.