Automated debris imaging system: calibrated global monitoring of floating marine litter using ship-based cameras and deep learning

Abstract

Continuous and global observation techniques are required for a comprehensive understanding of the mass balance and the temporal trends of plastic pollution in the aquatic environment. Efforts to achieve large-scale automated monitoring of floating marine debris have accelerated in recent years. For example, satellite-based mapping of floating marine debris in coastal regions has made substantial progress, aided by studies on the use of multispectral, hyperspectral, thermal infrared, polarimetry, synthetic aperture radar, and other methods. Satellite-based methods have proven challenging for the open ocean, where floating marine debris is too dispersed. Specifically, large floating macroplastics (>50 cm) constitute a significant part of the ocean plastic mass budget, but are poorly quantifiable even by classical methods, such as surface trawls and visual observations. Here, we demonstrate further refinements and a first-ever global campaign of the recently introduced Automated Debris Imaging System (ADIS). This technology employs optical imagery and deep learning to detect and classify floating macroplastics along vessel transects. For every detection, a geo-tagged snippet is created, allowing for human validation and filtering of false positives. Since the initial proof-of-concept, we have compiled a global dataset, comprising over 27 million images (165 TB), spanning 14 554 km2 over the north and south Pacific, north and south Atlantic, and Indian Ocean. In total, we identified more than 20 000 floating macroplastic items. Crucially, this study includes validation of the ADIS approach using concurrent surface net-trawl samples, enabling an assessment of detection accuracy and precision. These findings represent a major step forward in the development of scalable, high-resolution monitoring tools for marine plastic pollution in the open ocean.