Detection of Submerged Targets Beyond Eyes' Observation Using Satellite Lidar and Multispectral Data

Detecting submerged targets in shallow waters on satellite platforms remains a challenge, as the optical spectral information may be significantly distorted by the absorption and scattering effects of the water column, rendering it largely independent of the target itself. In this study, we propose a new framework as the bathymetry-informed target extraction (BITE), which integrates the active spaceborne lidar (ICESat-2) data and passive multispectral imagery (Beijing-3) to achieve satellite-based detection of submerged targets. By using the Satellite-Derived Bathymetry (SDB) model, we convert the complex multispectral information into relative depth data. Through this transformation, the challenging problem of distorted color domain image segmentation is converted into a problem of depth anomaly detection. Then, a generic target detection model is applied to this derived bathymetric map for automated and training-free target identification. The method is validated on submerged stone weirs in the Penghu Islands, which indicates significant improvements in target recognition rate and reliability compared to direct color-based methods. This approach promises rapid and large-scale surveys of submerged targets in shallow waters, offering an alternative solution to in-situsurveys such as shipborne sonars.