Hasan Zidni, Karisma Trinanda Putra, Nor Salwa Damanhuri, Sarwo Pranoto, Deri Aditya Romadhon, Rustam Asnawi
Accurate measurement of egg dimensions is essential in modern poultry farming for effective grading, sorting, and quality control. However, traditional manual methods are time-consuming and susceptible to inconsistency, limiting their scalability in large-scale operations. This study presents a hybrid approach that combines YOLOv8-based object detection with geometric analysis to automatically estimate egg lengths from digital images. A custom YOLOv8n model was trained on annotated egg datasets and evaluated for detection performance, achieving 0.87 mAP @ 0.5, 0.75 mAP @ 0.5: 0.95,0.88 precision, and 0.86 recall. Detected bounding boxes were converted from pixels to centimetres using a calibrated pixel-to-length ratio. The estimated dimensions were compared with ground-truth measurements taken manually using callipers. Across multiple test samples, the proposed method achieved a mean absolute error (MAE) of 0.056 cm and an average relative error of 1.29%, demonstrating high estimation accuracy. These results demonstrate both high estimation accuracy and the system's potential to reduce measurement time and human subjectivity, supporting scalable and contactless egg size classification in smart farming. © 2025 IEEE.
Universitas Negeri Yogyakarta, Departement Of Electrical Engineering Education, Yogyakarta, Indonesia; Universitas Muhammadiyah Bantul, Department of Electrical Engineering, Yogyakarta, Indonesia; University Teknologi Mara, Faculty of Electrical Engineering, Penang, Malaysia
