A marine ship detection method for super-resolution SAR images based on hierarchical multi-scale Mask R-CNN
Synthetic aperture radar (SAR) images have all-weather observation capabilities and are crucial in ocean surveillance and maritime ship detection. However, their inherent low resolution, scattered noise, and complex background interference severely limit the accuracy of target detection. This paper...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Frontiers Media S.A.
2025-07-01
|
| Series: | Frontiers in Marine Science |
| Subjects: | |
| Online Access: | https://www.frontiersin.org/articles/10.3389/fmars.2025.1574991/full |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Synthetic aperture radar (SAR) images have all-weather observation capabilities and are crucial in ocean surveillance and maritime ship detection. However, their inherent low resolution, scattered noise, and complex background interference severely limit the accuracy of target detection. This paper proposes an innovative framework that integrates super-resolution reconstruction and multi-scale maritime ship detection to improve the accuracy of marine ship detection. Firstly, a TaylorGAN super-resolution network is designed, and the TaylorShift attention mechanism is introduced to enhance the generator’s ability to restore the edge and texture details of the ship. The Taylor series approximation is combined to optimize the attention calculation, and a multi-scale discriminator module is designed to improve global consistency. Secondly, a hierarchical multi-scale Mask R-CNN (HMS-MRCNN) detection method is proposed, which significantly improves the multi-scale maritime ship detection problem through the cross-layer fusion of shallow features (small targets) and deep features (large targets). Experiments on SAR datasets show that TaylorGAN has achieved significant improvements in both peak signal-to-noise ratio and structural similarity indicators, outperforming the baseline model. After adding super-resolution reconstruction, the average precision and recall of HMS-MRCNN are also greatly improved. |
|---|---|
| ISSN: | 2296-7745 |