Lightweight vision models for egg fertility detection
Loading...
Date
2025-12-03
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Engineering, Technology & Applied Science Research
Abstract
The Philippine chicken industry relies heavily on effective egg fertility detection to sustain its growth. Traditional manual candling techniques are prone to human error and inefficiency; thus, there is a need for automation. While deep learning models such as Convolutional Neural Networks (CNNs) and YOLO have shown great potential for real-time object detection, a noteworthy gap in prior research is their application to native Philippine chickens due to a lack of publicly available image datasets. The current study fills this gap by developing and optimizing a YOLOv11 model for egg fertility detection in native Philippine chickens. A grid search was implemented to tune key parameters, such as the learning rate, optimizer (SGD, Adam, RMSProp), and weight decay, to improve the detector. Overall, careful tuning increases the model's performance. The best configuration used SGD with a 0.01 learning rate and a 0.0001 weight decay. The tuned model outperformed the baseline model in terms of mAP50-95, precision, and recall, while having faster inference speed. This focused tuning makes YOLOv11 more reliable in detecting the fertility of chicken eggs.
Description
Keywords
Chicken egg fertility, Computer vision, Deep learning, YOLO, YOLOv11, Object detection, Hyperparameter optimization (Grid search), Egg candling automation, Native Philippine chickens, Hatchery automation
Citation
Flores, E. J. (2025). Lightweight vision models for egg fertility detection. Engineering, Technology & Applied Science Research, 16(1), 31618-31623. https://doi.org/10.48084/etasr.15693