# Detection of bumblebee behaviour around a nest box using AI-based image analysis

## TL;DR

This paper uses AI to automatically count bumblebees entering and exiting a nest box in strawberry greenhouses, improving monitoring efficiency.

## Contribution

A novel algorithm combining YOLO with a crossing-counting method improves bumblebee behavior monitoring accuracy.

## Key findings

- The proposed method outperformed YOLO-only in accuracy, precision, and F1 score.
- Automated counting of bumblebee entry and exit behavior was successfully achieved.
- The method supports effective monitoring of pollinator behavior in controlled environments.

## Abstract

Insect pollinators, such as bumblebees, are commonly used to facilitate pollination in strawberry greenhouses. To ensure effective pollination, this study monitored the entry and exit behaviour of bumblebees around the nest box using an industrial camera. Video footage was captured, and a virtual cube-shaped frame was positioned at the entrance of the nest box. Bumblebee detection and counting within this virtual frame were performed using two methods: a) YOLO only and b) YOLO with a simple algorithm. In the algorithm-based method, if a bumblebee crossed the virtual frame an odd number of times, it was classified as having entered or exited the nest box. Conversely, an even number of crossings indicated that the bumblebee had either turned back or re-entered without exiting.

The proposed method allowed for the automated counting of bumblebees entering and exiting a nest box.

Compared to the YOLO-only method, the proposed method significantly improved performance metrics, including accuracy, precision, and F1 score.

The proposed method can effectively support the monitoring of pollinator behaviour in strawberry greenhouses.

Image, graphical abstract