Performance Evaluation of YOLO variants on Marine Trash Images: A Comparative Study of YOLOv5, YOLOv7, YOLOv8, and TinyYOLO

Despite the growing efforts to tackle the issue of marine pollution, the accurate and effective detection of oceanic litter remains a chronic difficulty. The development of a reliable deep learning-based system is vital for accurate and rapid marine trash detection in order to address these issues. This study delves into the comparison of the YOLO (You Only Look Once) variants, YOLOv4 tiny, YOLOv5, YOLOv7, and YOLOv8 so as to identify the best model for underwater plastic waste detection. The results of the experiment show that YOLOv5 has the highest mAP@50 at 0.92 and YOLOv5 also has a higher recall and F1-score compared to the other three models. Our findings also revealed that YOLOv7 had the smallest mAP@50, measuring at 0.847.