Sonar

Sonar Imaging

Join the Competition Contribute to Benchmark

Standard reconstruction benchmark — forward model perfectly known, no calibration needed. Score = 0.5 × clip((PSNR−15)/30, 0, 1) + 0.5 × SSIM

# Method Score PSNR (dB) SSIM Source
🥇 AcousticFormer 0.774 32.91 0.952 ✓ Certified Acoustic imaging transformer, 2024
🥈 SonarNet 0.667 28.37 0.888 ✓ Certified Underwater imaging CNN, 2022
🥉 MVDR/Capon 0.522 23.65 0.756 ✓ Certified Capon, Proc. IEEE 1969
4 DAS 0.470 22.23 0.700 ✓ Certified Van Trees, Array Processing, 2002

Dataset: PWM Benchmark (4 algorithms)

Blind Reconstruction Challenge — forward model has unknown mismatch, must calibrate from data. Score = 0.4 × PSNR_norm + 0.4 × SSIM + 0.2 × (1 − ‖y − Ĥx̂‖/‖y‖)

# Method Overall Score Public
PSNR / SSIM
 Dev
PSNR / SSIM
 Hidden
PSNR / SSIM
 Trust Source
🥇 AcousticFormer + gradient 0.724
0.780
31.9 dB / 0.942
0.711
28.35 dB / 0.888
0.682
26.83 dB / 0.854
✓ Certified Acoustic imaging transformer, 2024
🥈 SonarNet + gradient 0.594
0.696
26.62 dB / 0.849
0.562
21.72 dB / 0.678
0.525
20.4 dB / 0.618
✓ Certified Underwater imaging CNN, 2022
🥉 MVDR/Capon + gradient 0.501
0.563
21.82 dB / 0.682
0.488
19.2 dB / 0.560
0.453
18.84 dB / 0.542
✓ Certified Capon, Proc. IEEE 1969
4 DAS + gradient 0.481
0.524
20.43 dB / 0.619
0.483
19.43 dB / 0.571
0.436
17.61 dB / 0.481
✓ Certified Van Trees, Array Processing, 2002

Complete score requires all 3 tiers (Public + Dev + Hidden).

Join the competition →
Scoring: 0.4 × PSNR_norm + 0.4 × SSIM + 0.2 × (1 − ‖y − Ĥx̂‖/‖y‖) PSNR 40% · SSIM 40% · Consistency 20%
Public 3 scenes

Full-access development tier with all data visible.

What you get & how to use

What you get: Measurements (y), ideal forward operator (H), spec ranges, ground truth (x_true), and true mismatch spec.

How to use: Load HDF5 → compare reconstruction vs x_true → check consistency → iterate.

What to submit: Reconstructed signals (x_hat) and corrected spec as HDF5.

Public Leaderboard
# Method Score PSNR SSIM
1 AcousticFormer + gradient 0.780 31.9 0.942
2 SonarNet + gradient 0.696 26.62 0.849
3 MVDR/Capon + gradient 0.563 21.82 0.682
4 DAS + gradient 0.524 20.43 0.619
Spec Ranges (3 parameters)
Parameter Min Max Unit
sound_speed_profile -5.0 10.0 m/s
multipath -2.0 4.0
array_calibration -3.0 6.0 %
View Details →
Dev 3 scenes

Blind evaluation tier — no ground truth available.

What you get & how to use

What you get: Measurements (y), ideal forward operator (H), and spec ranges only.

How to use: Apply your pipeline from the Public tier. Use consistency as self-check.

What to submit: Reconstructed signals and corrected spec. Scored server-side.

Dev Leaderboard
# Method Score PSNR SSIM
1 AcousticFormer + gradient 0.711 28.35 0.888
2 SonarNet + gradient 0.562 21.72 0.678
3 MVDR/Capon + gradient 0.488 19.2 0.56
4 DAS + gradient 0.483 19.43 0.571
Spec Ranges (3 parameters)
Parameter Min Max Unit
sound_speed_profile -6.0 9.0 m/s
multipath -2.4 3.6
array_calibration -3.6 5.4 %
Submit Reconstruction View Details →
Hidden 3 scenes

Fully blind server-side evaluation — no data download.

What you get & how to use

What you get: No data downloadable. Algorithm runs server-side on hidden measurements.

How to use: Package algorithm as Docker container / Python script. Submit via link.

What to submit: Containerized algorithm accepting y + H, outputting x_hat + corrected spec.

Hidden Leaderboard
# Method Score PSNR SSIM
1 AcousticFormer + gradient 0.682 26.83 0.854
2 SonarNet + gradient 0.525 20.4 0.618
3 MVDR/Capon + gradient 0.453 18.84 0.542
4 DAS + gradient 0.436 17.61 0.481
Spec Ranges (3 parameters)
Parameter Min Max Unit
sound_speed_profile -3.5 11.5 m/s
multipath -1.4 4.6
array_calibration -2.1 6.9 %
Submit Algorithm View Details →

Blind Reconstruction Challenge

Challenge

Given measurements with unknown mismatch and spec ranges (not exact params), reconstruct the original signal. A method must be evaluated on all three tiers for a complete score. Scored on a composite metric: 0.4 × PSNR_norm + 0.4 × SSIM + 0.2 × (1 − ‖y − Ĥx̂‖/‖y‖).

Input

Measurements y, ideal forward model H, spec ranges

Output

Reconstructed signal x̂

About the Imaging Modality

Side-scan sonar maps the seabed by transmitting acoustic pulses perpendicular to the survey vessel's track and recording the backscattered energy as a function of time (range). The along-track resolution is determined by the beam width, while the across-track resolution comes from the pulse length. The sonar image is a 2D acoustic backscatter map where intensity encodes seabed roughness, composition, and the presence of objects. Acoustic shadows behind elevated objects provide height information. Challenges include multipath reflections, variable sound speed profile, and non-uniform ensonification.

Principle

Sonar imaging uses acoustic waves (typically 50 kHz to 1 MHz) to image underwater scenes. Active sonar transmits a sound pulse and records the echoes from the seabed, objects, or water column. The propagation speed in water (~1500 m/s, varying with temperature, salinity, and pressure) determines the time-to-distance relationship. Side-scan sonar and multibeam bathymetry produce 2-D and 3-D maps of the underwater environment.

How to Build the System

For side-scan sonar: mount a towfish with two transducer arrays (port and starboard) that ensonify a swath perpendicular to the survey track. For multibeam: mount a hull-mounted array (e.g., Kongsberg EM2040, 200-400 kHz). Sound velocity profiler (SVP) measurements are essential for ray-tracing corrections. Integrate with GNSS positioning and motion reference unit (MRU) for heave, pitch, and roll compensation.

Common Reconstruction Algorithms

  • Beamforming (delay-and-sum for multibeam sonar)
  • Synthetic aperture sonar (SAS) processing for enhanced azimuth resolution
  • Bottom detection and bathymetric surface extraction
  • Acoustic backscatter classification for seabed characterization
  • Deep-learning object detection for mine countermeasures or marine archaeology

Common Mistakes

  • Incorrect sound velocity profile causing depth and position errors
  • Multipath reflections (surface bounce, bottom bounce) creating ghost targets
  • Nadir gap (directly beneath the sonar) with no acoustic coverage
  • Motion artifacts from ship heave/pitch/roll not compensated
  • Side-lobe artifacts creating false targets near strong reflectors

How to Avoid Mistakes

  • Measure SVP at the survey site; update periodically during long surveys
  • Use multiple-return filtering and angle-based discrimination to remove multipath
  • Overlap adjacent swaths to fill the nadir gap; use a vertical beam sounder
  • Apply real-time MRU data for heave, pitch, and roll correction of depth measurements
  • Use advanced beamforming (CAPON, MVDR) to suppress side-lobe responses

Forward-Model Mismatch Cases

  • The widefield fallback produces a 2D (64,64) image, but sonar acquires 1D time-domain acoustic echo signals per beam — output shape reflects beamformed acoustic returns, not a spatial image
  • Sonar measurement involves acoustic wave propagation in water (c~1500 m/s, varying with temperature/salinity/pressure) with range-dependent attenuation and multipath — the optical-domain widefield blur has no connection to underwater acoustics

How to Correct the Mismatch

  • Use the sonar operator that models acoustic pulse transmission, seabed/target reflection, and receive beamforming: time-of-arrival encodes range, beam angle encodes bearing
  • Form sonar images using beamforming (delay-and-sum), SAS (synthetic aperture sonar) processing, or bathymetric extraction algorithms that require correct acoustic echo data format

Experimental Setup — Signal Chain

Experimental setup diagram for Sonar Imaging

Experimental Setup

Instrument: EdgeTech 4125 / Klein 3000 / Kongsberg EM 2040
Frequency Khz: 900
Range M: 100
Resolution M: 0.1
Swath M: 200
Platform: AUV / towed body
Application: seabed mapping / mine detection

Key References

  • Blondel, 'The Handbook of Sidescan Sonar', Springer (2009)

Canonical Datasets

  • UATD underwater acoustic target detection dataset
  • S3Simulator synthetic sonar (2024)

Spec DAG — Forward Model Pipeline

P(acoustic) → Σ_t → D(g, η₂)

P Acoustic Propagation (acoustic)
Σ Echo Integration (t)
D Hydrophone Array (g, η₂)

Mismatch Parameters

Symbol Parameter Description Nominal Perturbed
Δc(z) sound_speed_profile Sound speed profile error (m/s) 0 5.0
n_mp multipath Number of unmodeled multipaths 0 2
Δa array_calibration Array element calibration error (%) 0 3.0

Credits System

40%
Platform Profit Pool
Revenue allocated to benchmark rewards
30%
Winner Share
Top algorithm receives from pool
$100
Min Withdrawal
Minimum payout threshold
Spec Primitives Reference (11 primitives)
P Propagation

Free-space or medium propagation kernel (Fresnel, Rayleigh-Sommerfeld).

M Mask / Modulation

Spatial or spatio-temporal amplitude modulation (coded aperture, SLM pattern).

Π Projection

Geometric projection operator (Radon transform, fan-beam, cone-beam).

F Fourier Sampling

Sampling in the Fourier / k-space domain (MRI, ptychography).

C Convolution

Shift-invariant convolution with a point-spread function (PSF).

Σ Summation / Integration

Summation along a physical dimension (spectral, temporal, angular).

D Detector

Sensor readout with gain g and noise model η (Gaussian, Poisson, mixed).

S Structured Illumination

Patterned illumination (block, Hadamard, random) applied to the scene.

W Wavelength Dispersion

Spectral dispersion element (prism, grating) with shift α and aperture a.

R Rotation / Motion

Sample or gantry rotation (CT, electron tomography).

Λ Wavelength Selection

Spectral filter or monochromator selecting a wavelength band.