StreakNet-Arch: An Anti-Scattering Network-Based Architecture for Underwater Carrier LiDAR-Radar Imaging

Area of Science:

• Photonics and Optical Engineering
• Robotics and Automation
• Marine Technology

Background:

• Underwater imaging faces challenges with scattering, limiting depth and range.
• Existing methods for scatter suppression are often computationally intensive or less effective.
• Real-time 3D data acquisition in underwater environments is crucial for various applications.

Purpose of the Study:

• Introduce StreakNet-Arch, an end-to-end framework for real-time underwater 3D imaging.
• Enhance scatter suppression using Self-Attention and Double Branch Cross Attention (DBC-Attention).
• Validate the performance and efficiency of the Underwater Carrier LiDAR-Radar (UCLR) system.

Main Methods:

• Developed StreakNet-Arch, a binary-classification framework integrating Self-Attention and DBC-Attention.
• Utilized a self-developed Underwater Carrier LiDAR-Radar (UCLR) system.
• Conducted controlled water tank experiments and real-world trials in the South China Sea.

Main Results:

• StreakNet-Arch demonstrated superior scatter suppression compared to traditional bandpass filtering and MP networks.
• Achieved higher F1 scores than CNNs at similar model complexity.
• Real-time benchmarks showed constant average imaging time (54-84 ms), outperforming conventional methods.
• Validated UCLR system in the South China Sea with 46mm error at 1,000m depth and 20m range.

Conclusions:

• StreakNet-Arch offers a robust and efficient solution for real-time underwater 3D imaging.
• The novel attention mechanisms significantly improve scatter suppression.
• The UCLR system demonstrates practical applicability and high accuracy in challenging marine environments.