O'Brien, Daniel Steven (2024) Automated Cattle Counting: Leveraging YOLOv8 for Accurate Object Detection in Feedlot Environments. [USQ Project]
![[img]](https://sear.unisq.edu.au/style/images/fileicons/text.png) |
Text (Project – redacted)
OBRIEN_D_DISSERTATION_Redacted.pdf Download (4MB) |
Abstract
Based on wanting to modernise current livestock production processes, this project investigated whether an artificial intelligence-driven model can accurately detect cattle from remotely captured Unmanned Aerial Vehicle (UAV) images. By combining recent developments in artificial intelligence-driven, real-time object detection with well-established and tested methods for gathering aerial images (i.e., UAVs fitted with high-resolution cameras), we obtained accurate information about cattle numbers in tightly held feedlot pens. Five different pre-trained machine learning models that differed in size and complexity (i.e., YOLOv8-s, a smaller, faster model designed for real-time applications where speed is crucial and YOLOv8-m, a medium-sized model that offered improved accuracy at the cost of slightly slower inference times), were fine-tuned to recognise images of individual cattle from the UAV images captured. Models also compared object identification using both a full image, high resolution dataset (4056x3040 pixels) with a tiled image dataset, where the original full image dataset was divided into smaller 640x640 pixel tiles. Results indicated that the YOLOv8-m_full model outperformed all other models tested (i.e., YOLOv8-_tiled; YOLOv8-s_full; YOLOv8-s_tiled; RoboFlow_custom) in metrics related to Precision, Mean Average Precision, and recall. This result highlighted that a Convolutional Neural Networks parameters such as the number of network layers, directly impacts the detection accuracy in complex datasets. The project outcomes also present an applied system for the autonomous monitoring (i.e., detect, count, identify) of cattle in intensive feedlot settings that can reduce the amount of manual labour, and associated costs, currently required to monitor cattle in extensive production environments.
|
Statistics for this ePrint Item |
| Item Type: | USQ Project |
|---|---|
| Item Status: | Live Archive |
| Faculty/School / Institute/Centre: | Current – Faculty of Health, Engineering and Sciences - School of Engineering (1 Jan 2022 -) |
| Supervisors: | Long, Derek |
| Qualification: | Bachelor of Engineering (Honours) (Electrical & Electronic) |
| Date Deposited: | 17 Mar 2026 03:25 |
| Last Modified: | 17 Mar 2026 03:25 |
| Uncontrolled Keywords: | livestock production; Unmanned Aerial Vehicle (UAV) images; monitoring |
| URI: | https://sear.unisq.edu.au/id/eprint/53145 |
Actions (login required)
 |
Archive Repository Staff Only |
