Sidescan Only Neural Bathymetry from Large-Scale Survey.

Area of Science:

• Marine Geophysics
• Robotics
• Computer Vision

Background:

• Sidescan sonar offers high resolution and wide coverage for underwater surveys.
• Bathymetric data is often unavailable, necessitating cost-effective acquisition methods.
• Unmanned Underwater Vehicles (UUVs) and Unmanned Surface Vehicles (USVs) are increasingly used for marine exploration.

Purpose of the Study:

• To develop and assess a method for reconstructing bathymetry using only sidescan sonar data.
• To demonstrate the feasibility of this method for large-scale surveys.
• To provide an efficient and cost-effective solution for bathymetric data acquisition.

Main Methods:

• Formulating bathymetry reconstruction as a global optimization problem.
• Utilizing a Sinusoidal Representation Network (SIREN) to model bathymetry.
• Jointly estimating albedo and beam profile using a Lambertian scattering model.
• Comparing reconstructed bathymetry with high-resolution multi-beam echo sounder (MBES) data.

Main Results:

• Achieved an error of 20 cm in bathymetry reconstruction from large-scale surveys.
• Demonstrated the effectiveness of the proposed method with high-accuracy positioning.
• Validated the approach against established multi-beam echo sounder data.

Conclusions:

• The proposed method effectively reconstructs bathymetry from sidescan sonar data.
• This technique offers a valuable, cost-efficient alternative when bathymetric data is unavailable.
• Potential applications include shallow-water mapping with GNSS-equipped surface vehicles and AUVs utilizing SLAM.