Low-frequency volumetric synthetic aperture sonar (SAS) can image objects buried in the seabed as three-dimensional data cubes, but the achievable detection performance is bounded by environmental physics: in-sediment attenuation, rough water–sediment interface scattering, gas bubbles in the sediment, and sensor resolution. The source paper (Williams & Brown, POMA 2020) introduced a fast physics-normalized detection algorithm for buried and proud objects in volumetric SAS data and explicitly announced as ongoing/future work a dedicated convolutional-neural-network (CNN) classification stage to reduce false alarms; the fundamental detection limits imposed by these unavoidable environmental factors — which the paper argued cap detection capability below unity regardless of the algorithm employed — were discussed qualitatively but left unquantified. The problem is to deliver a physics-based, data-validated quantification of the fundamental detection limits and, within those limits, a CNN-based classification stage that reduces the false-alarm rate relative to detection alone.

Background

Synthetic aperture sonar (SAS) is the underwater analogue of synthetic aperture radar: a sonar moving along a straight track coherently combines echoes to synthesize a long array, achieving high along-track image resolution independent of range. Volumetric (3D) SAS extends this by using vertically distributed receiver arrays so that the returned energy is focused into a 3D voxel cube covering not only the seafloor but the sediment volume beneath it. A representative system is the Sediment Volume Search Sonar (SVSS), a towed low-frequency wideband system built to find unexploded ordnance and mines that are proud (resting on the seabed), partially buried, or fully buried. Because only sufficiently low acoustic frequencies penetrate the seabed, and because the returning signals traverse the sediment twice, detection of buried objects is governed by environmental factors the operator cannot remove: (i) in-sediment attenuation, which grows with frequency and burial depth; (ii) scattering at the rough water–sediment interface, which redirects energy on transmission and return; (iii) free gas bubbles trapped in muddy sediments, which produce 'acoustic turbidity' that can almost completely block sub-bottom penetration; and (iv) the sonar's own range/cross-track/vertical resolution and array geometry. In the measured data cubes, buried targets typically appear as weak localized volumes of acoustic return embedded in a cluttered, speckled sediment background, so a practical processing pipeline first runs a detection stage that finds candidate volumes, and then a classification stage that decides which candidates are man-made, since detection alone produces many false alarms. The source paper of this problem (Williams and Brown, Proceedings of Meetings on Acoustics 40, 070002, 2020) proposed such a fast detection algorithm for buried and proud objects in 3D volumetric SAS data cubes, built around a physics-motivated data-normalization step, and demonstrated it on real data from an experimental volumetric sonar. The authors' stated future work was a dedicated classification stage based on convolutional neural networks (CNNs, learnable multi-layer classifiers that extract hierarchical features directly from image or volume data) to reduce false alarms, and they noted that observed detection performance varied with environmental conditions (water temperature, microbial activity, sediment type) while the fundamental limits on detection capability imposed by sediment attenuation, interface scattering, gas bubbles, and sensor resolution remained unquantified. In follow-up work from 2020 to 2026 the same research groups substantially built out the announced classification strand — learned tone mapping for 3D SAS automatic target recognition (2023) and a trainable volumetric Hadamard normalization layer for sub-bottom target classification (2025) — while the fundamental detection-limit quantification the source discussed but did not perform has remained open. Adjacent literature makes clear why this matters operationally: offshore unexploded-ordnance survey practice currently relies on site-specific empirical surveying and data-quality control rather than any physics-derived statement of how deep, in what sediment, an object can reliably be found.

Problem Statement

For low-frequency volumetric synthetic aperture sonar (SAS) imaging — three-dimensional acoustic imaging in which a moving sonar coherently synthesizes a long aperture and focuses the returned energy into a voxel cube covering the seabed and the sediment volume beneath it — as realized in systems like the Sediment Volume Search Sonar (SVSS) and the data cubes processed by Williams and Brown (2020), establish the achievable buried-object detection performance envelope and demonstrate a false-alarm-controlled pipeline within it. Concretely: (1) quantify the fundamental limits on detection of buried and proud man-made objects imposed by the unavoidable environmental factors named in the source — in-sediment attenuation, rough water–sediment interface scattering, gas bubbles in the sediment, and the relationship between sensor resolution and target dimensions — whose combined effect the source argued caps the upper limit on detection capability below unity regardless of the algorithm employed, with the quantification validated against measured volumetric SAS data or, where measured data are not accessible, against physics-based simulated data generated by the cited validated scattering simulators, and reported in the source's ROC-like performance units (detection performance versus false-alarm rate per unit volume); variables such as burial depth, sediment type, acoustic frequency, and survey geometry may serve as illustrative axes of the data and analysis rather than a mandatory parameter set, and the portion of any observed performance shortfall attributable to these physical factors rather than to algorithm choice should be made explicit; and (2) within those limits, complete the detection-plus-classification pipeline with the dedicated classification stage based on convolutional neural networks (CNNs) that the source announced, demonstrating, at a stated detection probability and in the same detection-versus-false-alarm-rate units, a false-alarm rate reduced relative to detection alone, where the classification stage is assessed within the limits quantified in (1) and as part of the full detection-plus-classification pipeline, so that reproducing the cited classification work alone does not resolve (2); the source's observation that detection performance varied considerably across location and across time (suggesting strong environmental dependence, e.g., water temperature and microbial activity) motivates robustness across environmental variability but is not imposed as a fixed characterization requirement. The common objective is a volumetric-SAS buried-target search capability whose performance is predictable from quantified environmental physics rather than established empirically site by site.

References

  1. David P. Williams and Daniel C. Brown, 'New target detection algorithms for volumetric synthetic aperture sonar data,' Proceedings of Meetings on Acoustics 40, 070002 (2020), DOI 10.1121/2.0001296, https://doi.org/10.1121/2.0001296
  2. Tom E. Blanford, Daniel C. Brown, Shawn F. Johnson, and Cale F. Brownstead, 'Sediment Volume Search Sonar,' 24th European Meeting of Environmental and Engineering Geophysics (EAGE), pp. 1-2 (2018), DOI 10.3997/2214-4609.201802463, https://doi.org/10.3997/2214-4609.201802463
  3. Gregory D. Vetaw, Benjamin Cowen, Daniel C. Brown, David P. Williams, and Suren Jayasuriya, 'Learning-Based Tone Mapping to Improve 3D SAS ATR,' IGARSS 2023 - 2023 IEEE International Geoscience and Remote Sensing Symposium, pp. 6995-6998 (2023), DOI 10.1109/IGARSS52108.2023.10281904, https://doi.org/10.1109/IGARSS52108.2023.10281904
  4. Gregory D. Vetaw, Benjamin Cowen, Daniel C. Brown, David P. Williams, and Suren Jayasuriya, 'Volumetric Hadamard Normalization for Synthetic Aperture Sonar Sub-Bottom Target Classification,' IEEE Journal of Oceanic Engineering 50, pp. 3024-3038 (2025), DOI 10.1109/JOE.2025.3590095, https://doi.org/10.1109/JOE.2025.3590095
  5. Chris Su, Gregory D. Vetaw, Suren Jayasuriya, Glen Uehara, and Andreas Spanias, 'Random Quantum Circuits as 3D Convolutional Kernels for 3D SAS ATR,' 2025 16th International Conference on Information, Intelligence, Systems and Applications (IISA), pp. 1-4 (2025), DOI 10.1109/IISA66859.2025.11311281, https://doi.org/10.1109/IISA66859.2025.11311281
  6. Jason Philtron, Daniel C. Brown, Shawn Johnson, Kyle S. Dalton, Geoff Goehle, and Thomas E. Blanford, 'Data-model comparison of elastic target scattering for a shallow water sonar system for buried UXO detection,' The Journal of the Acoustical Society of America 153, A300 (2023), DOI 10.1121/10.0018928, https://doi.org/10.1121/10.0018928
  7. Hannah R. Kurdila, Geoff Goehle, and Daniel C. Brown, 'Morphological Component Analysis of Long-Duration Ringdown from Elastic Objects Imaged with the Sediment Volume Search Sonar,' OCEANS 2022, Hampton Roads, pp. 1-6 (2022), DOI 10.1109/oceans47191.2022.9977385, https://doi.org/10.1109/oceans47191.2022.9977385
  8. Thomas E. Blanford and David P. Williams, 'Considerations for Modeling Backgrounds in Synthetic Aperture Sonar Imagery,' OCEANS 2025 - Great Lakes, pp. 1-8 (2025), DOI 10.23919/oceans59106.2025.11244967, https://doi.org/10.23919/oceans59106.2025.11244967
  9. Miah Miner, Thomas E. Blanford, and Suren Jayasuriya, 'Exploring effects of tone mapping on buried targets using synthetic aperture sonar,' The Journal of the Acoustical Society of America 159, A270 (2026), DOI 10.1121/10.0045673, https://doi.org/10.1121/10.0045673
  10. Lei Yue, Hong Liang, and Yule Chen, 'A Significant Feature Recognition Method Based on the Acoustic Scattering Echoes of Underwater Buried Targets,' IEEE Transactions on Instrumentation and Measurement 75, art. 6511917 (2026), DOI 10.1109/TIM.2026.3709440, https://doi.org/10.1109/TIM.2026.3709440
  11. Stanisław Hożyń, 'A Review of Underwater Mine Detection and Classification in Sonar Imagery,' Electronics 10(23), 2943 (2021), DOI 10.3390/electronics10232943, https://doi.org/10.3390/electronics10232943
  12. Daniel Wehner and Torsten Frey, 'Offshore Unexploded Ordnance Detection and Data Quality Control—A Guideline,' IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 15, pp. 7483-7498 (2022), DOI 10.1109/JSTARS.2022.3200144, https://doi.org/10.1109/JSTARS.2022.3200144