Answers are grounded in this paper and cite their sources.
Underwater sub-bottom acoustic imaging plays a critical role in the remote inspection of submarine pipelines, which are essential for offshore energy transmission and submarine communication networks. Automated interpretation of sub-bottom profiler (SBP) data remains challenging due to severe noise contamination and complex background reverberations that obscure hyperbolic signatures of buried pipelines. To address these issues, this article proposes a hyperbolic morphology-guided deep-learning model, termed Hyperbolic Morphological Feature Embedded You Only Look Once, which explicitly embeds the intrinsic hyperbolic geometric priors into the detection pipeline. Specifically, a dynamic snake convolution module is integrated into the backbone network to mitigate scattering noise during feature extraction. Subsequently, a foreground–background refinement awareness module, incorporating vertex localization, symmetry perception, and background suppression, is designed to accentuate pipeline-related structural features while attenuating seabed clutter. Furthermore, an adaptive keypoint localization loss is jointly optimized with the conventional bounding-box regression loss to enable coarse-to-fine target positioning. Extensive experiments on real-world SBP imagery demonstrate that the proposed method achieves a recall of 76.47% and an F1-score of 82.10%, improving over the compared methods on the evaluated real-world dataset. The results confirm that embedding hyperbolic morphological constraints into deep neural networks offers a promising solution for automated submarine infrastructure inspection.
No takes yet. Share an insight, caveat, or question.
Zhang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: