Detection of Underwater Marine Plastic Debris Using an Augmented Low Sample Size Dataset for Machine Vision System: A Deep Transfer Learning Approach

Abstract

Waste in aquatic environments devastates aquatic habitats and offers a tall environmental and economical risk. Machine Vision might play a role in resolving this issue by detecting and finally eliminating debris. Using an augmented low sample size from a publicly available collection of underwater plastic waste, this research employed a YOLOv3 deep-learning system to visually recognize debris in realistic underwater environments. The detection model has a training and validation accuracy of 98.026 % and 94.582 %, respectively, according to the study's findings, with an mAP value of 98.15%. With its effectiveness in detecting underwater plastic waste, the recommended model is suitable for a variety of machine vision systems. The system has a 100% testing accuracy, with detection per frame accuracy ranging from 60.59% to 98.89%.