<![CDATA[Although the application of computer vision in poultry management has been widely studied, research on the early identification of duck sex based on egg morphology using deep learning on mobile platforms remains limited. This study aimed to design, implement, and evaluate a YOLOv8-based mobile application for detecting duck sex from egg images in real time. The study employed a Research and Development (R&D) approach with a prototype development model. The initial dataset consisted of 543 images encompassing the male egg, female egg, and non-egg classes. Data were obtained through field observations, interviews with duck farmers, and image acquisition using a smartphone camera. The data were then augmented through horizontal and vertical flipping, rotation, and zooming, resulting in 1,053 images divided into 70% training data, 10% validation data, and 20% test data. The YOLOv8s model was trained for 100 epochs and converted into TensorFlow Lite format for on-device inference in a Flutter-based Android application. The results showed that the model achieved recall values of 95% for the female egg class, 97% for the male egg class, and 100% for the non-egg class. Functional testing using black-box testing showed that all major features functioned properly. Meanwhile, usability testing using the System Usability Scale (SUS) involving 20 respondents yielded an average score of 79.62, which was classified as Acceptable, Grade B, and Good. This study concludes that the integration of YOLOv8s, TensorFlow Lite, and Flutter produces a functional application with an acceptable level of usability. This application has the potential to support the practical early identification of duck sex using mobile devices.]]>
Copyrights © 2026
