Detecting Trachinotus Blochii with a YOLOv8 Model and Video Dataset: A Deep Learning Approach for Fisheries Monitoring

Abstract

This study develops a deep learning-based object detection model using YOLOv8 to detect Bawal Bintang/Star Pomfret fish (Trachinotus Blochii) in real time and with high accuracy from a video dataset captured in a 3x3x3 m3 Floating Net Cage (KJA) off the Karimunjawa coast. The video was taken using an Underwater Camera placed at a position 1.5 meters deep at KJA. This study aims to develop an automated fish monitoring system to detect Star Pomfret fish and improve efficiency and accuracy in fisheries resource management. The YOLOv8s model was retrained for 50 epochs using an annotated fish video dataset, split into 80% for training and 20% for testing and validation, with a frame size of 640x640 pixels. The evaluation results indicate that this model performs well, with a precision of 0.8608, a recall of 0.8575, an F1-score of 0.8592, and a mAP50 of 0.9304. This study demonstrates the potential of applying the YOLOv8 model to an automated fish monitoring system in KJA, which can help improve efficiency and accuracy in managing fishery resources in Karimunjawa and its surroundings.