Global Detection and Morphological Characterization of Seamounts With Weakly Supervised Deep Learning

Abstract Seamounts are submarine volcanic features that record the tectonic and magmatic evolution of Earth's interior, yet their global distribution remains poorly resolved due to sparse high‐resolution bathymetric coverage. We present a weakly supervised deep learning framework that integrates gravity and bathymetric data to enable global‐scale detection and morphological characterization of seamounts. By adopting a vision foundation model and incorporating iterative pseudo‐label refinement, our method overcomes annotation scarcity and reveals numerous previously undocumented features. Elliptical Gaussian fitting further constrains the geometric validity of detections and enables standardized morphological analysis. The resulting inventory reveals tectonically organized variations in seamount morphology, with broad edifices along fast‐spreading ridges and steeper forms near intraplate hotspots, reflecting geodynamic controls on volcanic construction. Our study provides a transferable solution for automated geophysical interpretation under sparse or uncertain supervision, with potential applicability to other Earth systems characterized by limited annotations.