Implementation of YOLOV11 for Multi-Object Detection of Frozen Skipjack Tuna Freshness

Ensuring fish freshness is a critical challenge in the modern fishing industry, particularly for high-value commodities such as Skipjack tuna. This paper proposes a real-time multi-object detection system using a custom YOLOv11 model to automatically classify the freshness quality of frozen and no frozen Skipjack tuna (Katsuwonus pelamis). Three variants of YOLOv11, namely YOLOv11S, YOLOv11M, and YOLOv11L, were trained and validated on annotated image datasets. YOLOv11S achieved a test accuracy of 88.4%, validation accuracy of 90.1%, and training accuracy of 88.4%. YOLOV11M achieved higher performance, with test accuracy of 92.9%, validation accuracy of 94.8%, and training accuracy of 85.7%. The best results were obtained from YOLOv11L, which achieved a test accuracy of 91.5%, a validation accuracy of 94.9%, and a training accuracy of 83.7<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup> These findings indicate that YOLOv11L offers the highest performance, while YOLOv11M and YOLOv11S also showed competitive results. The proposed YOLOv11-based approach demonstrates potential for real-time fish freshness detection in automated quality control systems, aiming to reduce post-harvest losses and improve Indonesia's global seafood competitiveness.