A deep learning application for real-time debris detection: underwater environment

Abstract

Recognizing the significance of accurately measuring and removing underwater waste is vital for safeguarding marine ecosystems and the environment. Measuring underwater waste is challenging due to factors like light reflection, absorption, dispersed particulates, and color distortion. Detecting and measuring floating and surface waste is comparatively straightforward. The presence of marine waste is detrimental to both the environment and human health, as microplastics from decomposed waste can enter the food chain. In light of current circumstances, addressing water contamination is crucial for environmental preservation. A significant concern in today’s society is the contamination of water bodies. The absence of standardized benchmarks and data standards poses challenges in comparing research efforts related to automatic waste identification in underwater environments. This article tackles the issues of identifying underwater waste or debris by thoroughly examining existing publicly available underwater waste datasets and evaluating Deep Learningbased waste detection algorithms for underwater environments. Image processing, deep learning, and trawling hold promise in implementing effective solutions. Examination of publicly available datasets in this study can support future research efforts to protect our ecosystem. It consolidates prior research, presenting the results of tests conducted on the provided datasets, aiming to establish a reproducible benchmark for waste detection using YOLOv8 as well as classify the garbage using transformers (ViT and Swin) and transfer learning (DenseNet, VGG16 ResNet and InceptionV3). Used ICRA19 dataset encompasses a range of categories of waste, including bio, plastic, and ROV. On the other hand, we used the Forward Looking Sonar Image (FLS) Marine Debris Dataset having 10 Debris categories. The technique of this study achieves a maximum average accuracy which 92.2%, indicating successful waste detection and identification in underwater settings.