Related Experiment Video
Updated: Jun 13, 2026
09:19
Measuring the Structure, Composition, and Change of Underwater Environments with Large-area Imaging
Published on: April 18, 2025
SW-Net: A Direction-Aware Deep Learning Model for Shipwreck Segmentation in Side-Scan Sonar Imagery
1The School of Architecture, Harbin Institute of Technology, Shenzhen 518055, China.
Sensors (Basel, Switzerland)
|June 12, 2026
Summary
This study introduces SW-Net, a novel deep learning model for shipwreck detection using side-scan sonar. SW-Net enhances automated segmentation accuracy in challenging underwater conditions.
Area of Science:
- Marine archaeology
- Geophysical remote sensing
- Computer vision
Background:
- Side-scan sonar is vital for underwater cultural heritage preservation, enabling large-scale shipwreck detection in turbid waters.
- Automated segmentation of submerged targets is challenging due to speckle noise and seabed reverberations obscuring features.
Purpose of the Study:
- To develop an advanced deep learning model for accurate automated segmentation of shipwrecks from side-scan sonar data.
- To improve shipwreck detection robustness in complex underwater environments.
Main Methods:
- Proposed SW-Net architecture incorporating multi-scale input and a Directional Filter Bank for feature extraction.
- Utilized a directional attention mechanism to enhance structural feature modulation and address intensity inversions.
- Evaluated performance on the AI4Shipwrecks dataset against seven other segmentation architectures.
Main Results:
- SW-Net achieved the highest intersection over union (39.43%) and F1-score (56.56%) among evaluated methods.
- Demonstrated superior robustness against complex seabed interference.
- Exhibited the lowest computational complexity with 4.01 million parameters.
Conclusions:
- SW-Net offers a practical and efficient solution for shipwreck detection.
- The model is suitable for resource-constrained autonomous underwater vehicles, advancing underwater cultural heritage preservation.