Application of YOLOv11s Algorithm to Automate Egg Counting in Poultry Farms

Abstract

Chicken eggs are a strategic food commodity and a globally affordable source of animal protein. In the industrial poultry farming sector, accurate daily production data is fundamental for evaluating biological performance, feed efficiency, and supply chain management. However, the conventional manual counting process is highly susceptible to human error caused by physical fatigue and dynamic environmental conditions, resulting in significant data inaccuracies. To address these inefficiencies, this study aims to develop a precise automated egg counting system by implementing the You Only Look Once (YOLO) deep learning algorithm version 11s. This study employed an experimental approach that included data collection, manual annotation, and extensive augmentation techniques to ensure the model's robustness to variations in lighting and spatial orientation. The model was trained using a specific egg dataset and evaluated against standard object detection performance metrics. The experimental results demonstrated superior performance, with the model achieving 97.80% Precision, 98.60% Recall, and 99.10% Mean Average Precision at intersection over union 50%. Furthermore, the F1-Score reached 98.20%, indicating a strong balance between detection accuracy and sensitivity. The system successfully detected and counted eggs in a real-time environment while maintaining stability across various visual scenarios. In conclusion, the proposed architecture proved to be a highly effective solution for automating production recording in poultry farms. These findings offer a practical transition towards digitized precision farming, significantly reducing operational inefficiencies associated with manual monitoring while enabling sound managerial decision-making.