<![CDATA[This study implements the YOLOv11 algorithm to detect five types of ornamental guppy fish through an Android-based application. The research background arises from the difficulty of manually distinguishing guppy varieties due to their complex color variations and patterns. The methodology includes dataset collection, labeling using Roboflow, image preprocessing and augmentation, training the YOLOv11n model, conversion to TensorFlow Lite, as well as real-time implementation and testing within the application. The training results demonstrate strong performance, achieving a Precision of 85.90%, Recall of 90.70%, mAP50 of 90.60%, mAP50–95 of 66.40%, and an F1-Score of 88.3%. Indirect testing on 250 test images produced per-class accuracy ranging from 94% to 98%. Direct real-time testing indicates that distance and fish orientation significantly influence confidence scores: at 10 cm the confidence reached 76.6% (straight) and 71.6% (turned), at 15 cm it reached 83% (straight) and 75.4% (turned), and at 20 cm it reached 72.8% (straight) and 47.4% (turned). The optimal performance was obtained at a distance of 15 cm. The application is capable of detecting objects within 1–3 seconds, making it suitable for real-time guppy identification. This study can be further improved by expanding the dataset, optimizing the model, and adding additional application features. Keywords: YOLOv11, Guppy Fish Identification, Object Detection, Image Processing, Android Application]]>
Copyrights © 2026
