Date

2023-1-24

Author

Has, Alper Emre

Metadata

Show full item record

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

Item Usage Stats

246
views

172
downloads

Cite This

Beekeeping is a valuable field for both scientific research and agriculture. While bee keepers have traditionally relied on close observation, often with the help of vets, this method can provide limited empirical results. The high cost of monitoring tools presents a challenge in the beekeeping industry. To address this challenge, we im plement an animal detecting and tracking algorithm as part of a framework that uti lizes low-cost cameras and low-power computational equipment. This leads to the development of a semi-automated Image Labeling Software (ILS) that reduces the workload of labeling for similar frameworks. Our ILS was used in the Beehive 4.0 project, which uses an artificial intelligence-based approach for observing honeybees (Apis mellifera) in a controlled environment. Using a low-resolution camera, we simultaneously record the test group and control group. We trained a standard YOLOv3 on our dataset and use this for object detec tion. To increase accuracy and reduce false positive detections, we also identified the contours of bee clusters. We then used k-means clustering to eliminate redundant detections. This study demonstrates that object detection and tracking on low-cost camera-recorded videos can be improved by using the proposed model in animal keeping. This will enhance labeling quality while reducing the labeling burden in reporting setups when the number of animals is known.

Subject Keywords

semi-automated image labeling systems, honeybee, object detection, object tracking, YOLOv3

URI

https://hdl.handle.net/11511/102115

Collections

Graduate School of Natural and Applied Sciences, Thesis