High-resolution bathymetry by deep-learning-based image superresolution.

Area of Science:

• Oceanography
• Geophysics
• Computer Science

Background:

• High-resolution bathymetric charts are crucial for ocean exploration and management.
• Current methods for creating these charts are often time-consuming and expensive due to extensive data acquisition requirements.

Purpose of the Study:

• To develop an efficient method for enhancing the resolution of bathymetric charts using deep learning.
• To reduce the cost and time associated with detailed seafloor mapping.

Main Methods:

• Gridded bathymetric data were treated as digital images.
• A deep-learning-based superresolution technique was employed to estimate high-resolution bathymetric images from low-resolution ones.
• The model automatically learned geometric features of the ocean floor to recover details.

Main Results:

• The deep-learning superresolution method significantly outperformed naive interpolation techniques.
• An average improvement of eight decibels in peak signal-to-noise ratio was observed.
• Qualitative and quantitative assessments confirmed the effectiveness of the proposed approach.

Conclusions:

• Deep-learning-based superresolution offers a powerful tool for enhancing bathymetric chart resolution.
• This technology can substantially decrease the required sea area measurements, accelerating global seafloor mapping.
• The method promises to make high-resolution bathymetric data more accessible and cost-effective.