<![CDATA[Catfish aquaculture in Indonesia faces an efficiency challenge in its fry counting process, which still relies on manual methods. This research aims to develop and evaluate a mobile application based on the Android operating system that implements the YOLOv5 algorithm for the automated, real-time detection and counting of catfish fry. The model was trained using an image dataset from Roboflow and integrated into an application developed with the Flutter framework. The model's performance was quantitatively assessed using Precision, Recall, and F1-Score metrics across three scenarios: normal, clustered (occlusion), and shadowed conditions. The test results show that the best performance was achieved under normal conditions, with an F1-Score of 0.949. Performance decreased when the fry was clustered (F1-Score of 0.874) due to object occlusion, and also under shadowed conditions (F1-Score of 0.786) because of false positive detections. These findings confirm the suitability of YOLOv5 for fry counting applications on mobile devices while also highlighting critical areas for improvement, particularly in handling lighting variations and overlapping objects.]]>
Copyrights © 2025
