Underwater Debris Detection Using Visual Images and YOLOv8n for Marine Pollution Monitoring

A Huge amount of solid waste is generated daily from human activities in residential, industrial, agricultural, or commercial areas, leading to seashore, floating, and underwater debris due to improper disposal of solid wastes, which can negatively impact aquatic flora and fauna. Identification of underwater debris would be a challenging task because of poor water transparency and poor quality of underwater images generated even with underwater vehicles enabled high-quality camera(s) under complex underwater environments. In this paper, we present a deep learning-based underwater debris detection method using different quality visual images and the YOLOv8n network. The trained YOLOv8n model is obtained by using 2000 images taken from the Trashcan dataset for classifying the image into debris and non-debris. The effectiveness of the YOLOv8n model is evaluated in terms of precision (PR), recall (RE), mean average precision (mAP), and computational load on a wide variety of untrained images taken from the trained database and completely untrained image database. Evaluation results demonstrate that the YOLOv8n-based model achieves an m $A$P of 86.01%, PR of 88%, and RE of 94.6% that outperforms existing YOLO versions (Tiny YOLO, YOLOv2, YOLOv3, YOLOv5s), faster region-based convolutional neural network (R-CNN), mask R-CNN, single stage detector (SSD) and You Only Look At CoefficienTs (YOLACT) based models. The YOLOv8n-based debris detection model had improvements of 11.39% precision, 50.16% recall, and 28.37% mAP as compared to the improved YOLOv5s model. The YOLOv8n model had a mean average precision improvement from 63.26% to 20.08% as compared to other YOLO models.