Automated Underwater Plastic Waste Detection and Volume Assessment through Instance Segmentation

Abstract

Plastic pollution in our oceans has become an urgent ecological issue as it endangers aquatic ecosystems and diminishes biodiversity. One problem with identifying plastic waste that is submerged in water is that underwater visibility is poor, with low-light penetration, poor colouration, and cluttering of underwater areas. Thus, this article proposes an intelligent system that detects plastic waste in water and identifies its volume and size based on the YOLOv8nseg instance segmentation model. The authors explain their approach of applying the SeaMarine underwater plastic segmentation dataset and preprocessing this dataset using the Roboflow platform. The system is capable of identifying various types of underwater debris and outputs segmentation masks for all detected objects. The system measures the volume of waste by using the segmentation masks developed from the detected regions. The authors provide a user-friendly system based on Gradio functions so that the user can effectively analyze underwater videos and visible waste. Thus, the paper presents a novel intelligent technology that brings various and complex tasks into a single piece of software and opens new horizons in the area of underwater environmental monitoring.