Real-time detection of submerged debris in aquatic ecosystems using YOLOv8

Abstract

The current circumstances, it is imperative to prioritize the implementation of effective strategies to mitigate the negative Water contamination is a prominent environmental concern in contemporary society. The lack of standardized benchmarks and measurements, as well as the absence of widely acknowledged standards for data, make it challenging to compare current research efforts in the field of automatic waste identification. This article addresses the aforementioned issues by offering a comprehensive examination of current underwater waste datasets and a concise yet insightful evaluation of existing waste detection algorithms based on Deep Learning. This article compiles and synthesizes prior research and presents the findings of the authors' own tests conducted on the provided datasets. The objective is to provide a first reproducible benchmark for detecting waste using YOLOv8. These datasets are collected from open-source datasets. They include comprehensive annotations that encompass all conceivable waste categories, such as bio, glass, metal and plastic, non-recyclable, other, paper, and unknown. The technique provided in this study demonstrates a maximum average accuracy of 92% on the dataset. The findings indicated that the enhanced model successfully achieved the detection and identification of waste present in underwater.