Physics-Informed Side-Scan Sonar Perception: Tackling Weak Targets and Sparse Debris via Geometric and Frequency...

Area of Science:

• Marine Robotics
• Artificial Intelligence
• Acoustic Signal Processing

Background:

• Side-scan sonar (SSS) is crucial for Autonomous Underwater Vehicle (AUV) search and rescue (SAR).
• Challenges include hydro-acoustic noise, discontinuous wreckage, and weak target signatures.
• Existing methods struggle with accurate detection in complex underwater environments.

Purpose of the Study:

• To develop an enhanced detection network, WPG-DetNet, for AUV-based marine SAR.
• To improve the accuracy and efficiency of identifying critical targets in challenging underwater conditions.
• To address limitations in current sonar-based detection systems.

Main Methods:

• Introduced a Wavelet-Embedded Residual Backbone (WERB) using Discrete Wavelet Transform (DWT) for noise reduction and edge preservation.
• Incorporated a Debris Graph Reasoning Module (D-GRM) to understand scattered wreckage contextually.
• Designed a Shadow-Aided Decoupling Head (SADH) with physics-informed loss for target-seabed differentiation.

Main Results:

• WPG-DetNet achieved 97.5% mAP50 and 96.9% Recall on the SCTD dataset.
• Outperformed Faster R-CNN by 12.8% in mAP50 and RT-DETR-R18 by 5.6% in mAP50:95.
• Demonstrated efficient real-time performance with 62.5 FPS and 16.8 M parameters.

Conclusions:

• WPG-DetNet effectively overcomes noise and discontinuity challenges in underwater SAR.
• The integrated modules provide robust perception and context-aware scene understanding.
• The framework offers a balanced solution for AUV operations, enhancing search and rescue capabilities.