Underwater bio-inspired optical navigation based on deep learning and ambient light polarization characteristics

Area of Science:

• Marine navigation
• Optical oceanography
• Robotics

Background:

• Traditional navigation methods fail underwater due to electromagnetic wave limitations.
• Natural underwater light polarization patterns offer stable, interference-resistant navigation cues.
• Existing methods lack precision and accumulate errors in submerged environments.

Purpose of the Study:

• To develop and validate a novel underwater navigation system utilizing natural light polarization.
• To leverage polarization patterns for accurate solar position and geographic coordinate estimation.
• To address the limitations of conventional navigation in marine settings.

Main Methods:

• Feature extraction of underwater polarization patterns (angle, degree, intensity).
• Application of a squeeze-and-excitation ResNet50 model for solar position estimation (zenith, azimuth).
• Utilizing compressed sensing for precise latitude and longitude prediction.

Main Results:

• Achieved low average errors: 0.189° (zenith), 0.158° (azimuth), 0.117° (latitude), 0.034° (longitude) for single locations.
• Demonstrated robust performance across multiple locations with errors of 0.622° (zenith), 0.295° (azimuth), 0.238° (latitude), 0.023° (longitude).
• Validated the efficacy of polarization patterns for reliable underwater navigation.

Conclusions:

• Natural underwater polarization patterns provide a viable and accurate navigation solution.
• The ResNet50 and compressed sensing approach effectively estimates solar position and location.
• This method offers a promising, low-error alternative for autonomous underwater vehicle navigation.