S2DDM: single-beam-to-DEM diffusion model for large-scale consistent seabed terrain restoration based on single-beam bathymetric data

Single-beam bathymetry, with its early development, has accumulated extensive data during ocean exploration, serving as a crucial source for global seabed topographic modeling. Although single-beam bathymetric historical data are widely distributed globally, they exhibit random and sparse discrete point clouds locally, limiting their ability to continuously represent seabed topography and geomorphology. Applying diffusion model theory, in this study, the Single-beam-to-DEM diffusion model (S2DDM) is proposed as a model for reconstructing complete seabed DEMs from sparse single-beam bathymetric data. The S2DDM uses single-beam bathymetric data as constraints and topographic feature lines as prior knowledge, guiding terrain prediction in unmeasured areas between survey lines. This approach generates semantically coherent and structurally clear topography within patches. Simultaneously, the S2DDM eliminates elevation discontinuities between adjacent patches through edge-matching error processing, enabling large-scale reconstruction. The proposed method is tested in multiple regions around the world. Compared with the baseline models, the S2DDM not only effectively improves the visual effect, but also significantly improves the elevation accuracy, and ultimately achieves a reduction in the RMSE by more than 44% and an increase in the PSNR by more than 5.17 dB. It can produce practical value in the task of seabed topography and geomorphology refinement modeling.