GEMİ TESPİTİ UYGULAMASINDA YOLOV8 VE YOLOV9 ALGORİTMALARININ PERFORMANS DEĞERLENDİRMESİ

Tekindemir B., Şenel F. A.

Uluslararası Sürdürülebilir Mühendislik ve Teknoloji Dergisi, cilt.8, sa.2, ss.192-199, 2024 (Hakemli Dergi)

Özet

The detection and classification of vessels represents a pivotal challenge in the domain of maritime surveillance and monitoring. Its applications encompass a diverse range of fields, including fisheries management, migrant monitoring, maritime rescue operations, and maritime warfare. The utilisation of remote sensing technologies for the tracking of ships is a consequence of the advantages they offer, including extensive coverage and low-cost accessibility. This study emphasizes the importance of human detection, counting and tracking of objects using computer vision and machine learning methods. In this study, the potential of YOLO architectures as a technology for rapid and precise ship detection and classification is explored. The YOLOv8 and YOLOv9 architectures were employed for the detection of ships utilising remote sensing techniques. This study compares the performance of the YOLOv8 and YOLOv9 architectures using a dataset, referred to as the "Ships in Google Earth" dataset, which consists of 1658 images for ship detection. The models were evaluated in terms of training and validation losses, precision, recall and average precision, and demonstrated a certain degree of success and learning speed during the training process. Both models were found to provide effective solutions for ship detection. However, the YOLOv9 model exhibited superior performance in terms of faster convergence and overall detection accuracy, particularly at the outset.