Marine pollution is a global concern, with vast amounts of suspended debris distributed throughout the water column beyond the well-known surface and seafloor debris. The density distribution of suspended debris across different depth layers reflects the spatial behavior of pollutants and is key to understanding the pollution process in low-visibility environments. This study proposes an automated perception framework that integrates high-resolution Forward-Looking Sonar (FLS) with optical flow estimation to enable real-time detection, height estimation, and density modeling of suspended debris. The proposed method first detects potential debris targets by analyzing strong backscattering features and motion cues in sonar images. It then accurately estimates the relative height of these targets above the seafloor using a sonar imaging geometry model. Leveraging the navigation data of the underwater observation platform and the temporal accumulation of successive image frames, the system normalizes and counts the number of detected targets within each depth layer, constructing a layered density distribution map to reveal vertical aggregation patterns. This method features real-time capability, does not rely on manual annotation, and offers strong physical interpretability. It can be integrated into unmanned underwater platforms for large-scale surveys. Underwater experiments demonstrate that the proposed framework exhibits robust performance and provides a novel perspective for 3-D behavioral modeling of marine debris.
Zhou et al. (Wed,) studied this question.