Visual Transformers for Scalable Detection and Mapping of Ocean Plastic Debris

The escalating challenge of ocean plastic pollution necessitates innovative detection and cleanup strategies. Traditional methods are hindered by high costs and the inability to provide real-time data, particularly for submerged plastics. This study proposes a novel solution leveraging Vision Transformers (ViTs) and low-cost hardware for onboard classification. A ViT model, trained on a comprehensive dataset of aquatic images, classifies plastic debris in real-time on a Raspberry Pi 4 with a Camera Module 2. Integrated with a Video Plankton Recorder (VPR) system and GPS, it maps plastic distribution, including submerged plastics. Installing this setup on cargo vessels, which traverse oceans regularly, reduces detection costs and enables continuous monitoring. This scalable, cost-effective approach enhances ocean plastic monitoring, facilitating targeted cleanup efforts and supporting environmental conservation. By combining deep learning with accessible hardware, this solution advances ocean cleanup efforts and protects marine biodiversity.