The MARVEL Dataset: Marine Debris Analysis and Recognition for Vision-Based Exploration

Aquatic ecosystems are at risk from marine debris, but current databases are not diverse enough to detect it accurately in the real world. We fill this void by presenting the MARVEL (Marine Debris Analysis and Recognition for Vision-Based Exploration) dataset, which is intended to categorize various forms of waterborne garbage, including shopping bags, plastic particles, rotting fruits, and thermocol. The distinct features of marine debris are not captured by existing datasets such as TrashNet and MJU-garbage, which mostly concentrate on general garbage. In order to guarantee precise classification for reliable model training, our dataset contains carefully annotated photos using the LabelMe tool. Vision-based detection systems are improved by the MARVEL dataset, increasing their efficacy in practical applications. It enables improvements in marine pollution control and aids in international efforts to protect aquatic ecosystems by addressing a crucial research gap.