Thermal Imaging and Advanced Deep Learning for Automated Broiler Detection and Counting

Abdalshefie Abuhussein, Mohamed Fawzi; El-Soaly, Ibrahim S.; Elmessery, Wael M.; Abd El-Wahhab, Gomaa G.

doi:10.21608/azeng.2024.282011.1013

	Thermal Imaging and Advanced Deep Learning for Automated Broiler Detection and Counting
Al-Azhar Journal of Agricultural Engineering
Volume 7, Issue 1, June 2024 PDF (1.37 MB)
Document Type: Original Article
DOI: 10.21608/azeng.2024.282011.1013
View on SCiNiTO
Authors
Mohamed Fawzi Abdalshefie Abuhussein ¹; Ibrahim S. El-Soaly²; Wael M. Elmessery³; Gomaa G. Abd El-Wahhab²
¹Department of Agricultural Constructions Engineering and Environmental Control, Faculty of Agricultural Engineering, Al-Azhar University, Cairo, Egypt
²Department of Agricultural Constructions Engineering and Environmental Control, Faculty of Agricultural Engineering, Al-Azhar University, Cairo, Egypt
³Department of Agricultural Engineering, Faculty of Kafr Elsheikh University, Kafr Elsheikh, Egypt
Abstract
Poultry farming plays a vital role in meeting the rising global demand for animal protein. However, traditional techniques for monitoring broiler chickens have limitations. Manual broiler counting is time-consuming and prone to errors. This study explores using thermal imaging and advanced deep learning for automated broiler detection and counting. A dataset of 5000 thermal image frames of 2000 chickens was created for training, validation, and testing. Two deep learning models, YOLOv7 and YOLOv8, were compared. Thermal image features were extracted using these models to capture broiler-related thermal patterns. The study addresses the challenge of automating broiler detection and counting with high accuracy and efficiency. YOLOv8 outperformed YOLOv7, achieving significantly higher mean Average Precision (mAP) of 95% compared to 85%. Faster convergence within 20 epochs compared to 60 epochs for YOLOv7. In addition, YOLOv8 exhibited lower error rates of 2% vs. 5% for broiler counting tasks. This research demonstrates the effectiveness of YOLOv8 for real-time precision agriculture applica-tions using thermal imaging and deep learning for poultry monitoring. The findings pave the way for implementing automated broiler detection and counting systems in poultry farms, improving efficiency and data accuracy.
Keywords
Thermography; intensive poultry houses; YOLO-based object detection


Statistics Article View: 279 PDF Download: 382