Marine Waste Detection Using YOLOv3: A Real- Time Deep Learning Approach

Abstract

In the domain of marine conservation, the urgency to address the increasing crisis of marine waste has never been more pronounced. This study presents the implementation and evaluation of a YOLOv3-based deep learning model for underwater waste detection. The system leverages pre-trained YOLOv3 weights and an underwater dataset annotated for underwater debris classification. By fine- tuning the model and applying Non-Maxima Suppression (NMS) for precise bounding box generation, the study achieved high detection accuracy and real-time processing. The results demonstrate the capability of YOLOv3 in identifying underwater waste under challenging conditions, including varying lighting and water clarity. The proposed system provides an efficient, scalable, and adaptable solution for environmental monitoring and conservation efforts.