<![CDATA[Eggs are an important source of animal protein and are widely consumed by the public. However, quality issues such as cracked or broken eggs are still frequently encountered during distribution and storage. Egg quality sorting has been largely done manually, making it prone to human error, time-consuming, and inconsistent. This study aims to develop a deep learning-based egg quality classification system with a transfer learning approach using the MobileNetV2 architecture that is efficient for devices with limited computing capacity. The research method involves acquiring egg image datasets (good and broken), preprocessing data with normalization and augmentation, designing a MobileNetV2 model, conducting two-stage training (feature extraction and fine-tuning), and evaluating model performance. Implementation was carried out both in the development environment and on a Jetson Nano edge computing device to test real-time application. The results showed that training with fine-tuning increased classification accuracy to 92% with an average precision, recall, and F1-score of 0.95. Confusion matrix evaluation demonstrated the model's ability to distinguish egg classes well, although there were still small errors in the classification of "good" eggs. Implementation on the Jetson Nano demonstrated relatively fast inference times (50–70 ms) with low resource consumption, demonstrating the system's applicability at both farm and small-to-medium scale distribution. This research successfully presented an accurate, lightweight, and practically implementable egg classification model as a first step towards automating the egg sorting process in the livestock industry. Implementasi pada Jetson Nano menunjukkan waktu inferensi yang relatif cepat (50–70 ms) dengan konsumsi sumber daya yang rendah, menunjukkan penerapan sistem pada pertanian dan distribusi skala kecil hingga menengah.]]>
Copyrights © 2026
