Deep Learning Approach for Marine Plastic Waste Detection in Autonomous Robots

Abstract

Marine plastic pollution is a major global environmental issue that affects marine ecosystems, biodiversity, and human health. Each year, large amounts of plastic waste enter oceans through rivers, coastal activities, industrial discharge, and improper waste disposal. These plastics remain in the environment for long periods, harming marine life through ingestion, entanglement, and habitat damage. Over time, they break down into microplastics, which contaminate water and enter the food chain, posing long-term ecological and health risks. Traditional methods of monitoring marine plastic waste, such as manual inspection and satellite observation, are often slow, expensive, and limited in real-time effectiveness. To address these challenges, this research proposes a deep learning-based system combined with autonomous robotics for efficient plastic waste detection. The system uses cameras mounted on robots to capture real-time ocean images, which are analysed using Convolutional Neural Networks (CNNs) to detect and classify plastic items like bottles and bags. The methodology involves dataset collection, pre-processing, model training, and robotic integration. Techniques like image resizing, normalisation, and data augmentation improve model performance. Advanced object detection models such as YOLO, Faster R-CNN, and SSD are evaluated, with YOLO achieving up to 92% accuracy and strong performance even under challenging conditions like reflections and waves. This system enables real-time monitoring, reduces human effort, and supports large-scale ocean cleanup operations. It also allows continuous data collection for analysing pollution patterns and supports better environmental decision-making. While challenges such as microplastic detection and environmental variability remain, this approach offers a scalable and effective solution for tackling marine plastic pollution and promoting sustainable environmental conservation.