Semi supervised orientation estimation of honeybees

Date

2025-7-21

Author

Gündeğer, bilal yağız

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

6422
views

0
downloads

Cite This

Honeybees play a vital role in the environment, particularly in pollination, yet their populations are declining at an alarming rate. They interact not only within the colony but also extensively with their surroundings. Understanding their behavior is crucial for conservation efforts, and a key aspect of this is accurately estimating the pose of individual bees. Traditional methods require attaching physical markers, such as QR-coded tags, to the bees' bodies, which can be intrusive. Recent advances in deep learning–based computer vision have made it possible to estimate the pose of bees, including both their location and orientation within a coordinate system. However, such models are typically trained in a supervised manner and therefore require large amounts of labeled data, consisting of both input samples and their corresponding annotations. The reliance on extensive labeled datasets is a major limitation, as it demands significant annotation effort to achieve effective learning and robustness. Furthermore, these models often struggle to generalize when confronted with long-term data distribution shifts that are not captured in the training data. This thesis presents a novel orientation estimation model for honeybees that operates alongside existing detection systems. The approach assumes that reliable bee detections in the form of bounding boxes aligned with the coordinate axes are available as input and estimates the orientation of detected honeybees. It can be easily combined with various detection model without requiring modifications to the detection process itself since the orientation estimation can run as a separate module. The orientation estimation model has been trained using a semi-supervised approach, requiring only a small amount of labeled data for initial training, while automatically labeling the remaining data. To ensure long-term performance, we applied a lifelong learning strategy, enabling the system to adapt to dynamically changing data distributions. The overall system, which includes both the model and the semi-supervised pipeline, continuously labels incoming data and periodically retrains the model. In our experiments, the proposed method achieved a mean absolute error of approximately 16 degrees on the test dataset consisting of 60 frames captured from an observation hive at the University of Graz, Austria, demonstrating performance comparable to state-of-the-art approaches despite minimal labeled data. Its ability to maintain accuracy with only ~1-degree variation over long-term applications highlights its potential for supporting the study and monitoring of honeybee populations.

Subject Keywords

Markerless tracking, Orientation estimation, Lifelong learning, Honeybee, Behavior, Semi-supervised learning

URI

https://hdl.handle.net/11511/115538

Collections

Graduate School of Natural and Applied Sciences, Thesis