해양 생태계 모니터링을 위한 YOLOv8 기반 어류 탐지 AI 모델 개발 및 개인정보 보호 기법 적용

This study presents the development of a deep learning fish species classification model targeting eight representative fish species found along the coastal waters of South Korea, including the East Sea, West Sea, and South Sea. Traditional visual identification methods are time-consuming, labor-intensive, and often inaccurate under complex environmental conditions. To address these limitations, we propose an automated detection model based on the state-of-the-art object detection algorithm YOLOv8. A dataset of 2,339 images was collected under diverse lighting and background conditions to ensure reliability and variability. To safeguard personal information, a pre-trained face detection model was integrated to implement an automatic mosaic processing mechanism. Additionally, the LabelImg annotation tool was employed to generate precise bounding boxes, enhancing detection performance. The model was optimized through hyperparameter tuning and transfer learning in a GPU-based environment. Evaluation metrics including F1-score, the precision-recall curve, and the confusion matrix demonstrate the detection model's robustness and high accuracy across various settings. The proposed model holds significant potential for applications in fishery resource management, real-time monitoring, catch analysis, and ecosystem-change surveillance. In particular, it is expected to serve as a valuable foundation for sustainable fisheries policies and marine ecosystem conservation.