Scholarlink

<p>Ship detection represents a significant application of synthetic aperture radar (SAR) technology. Despite its advantages, SAR-based ship detection models face challenges due to unique imaging characteristics, including substantial scattering noise and arbitrary ship orientations. To address these issues, we propose Enhanced rotating ship Detection in SAR Images via Noise Suppression and Feature Amplification (NSFA-Net) for rotating SAR ship image detection. NSFA-Net introduces a frequency domain attention module to highlight foreground targets and suppress background noise, enabling the construction of an effective feature extraction network. An optimized aggregation network-GR is employed to mitigate redundant noise generated during feature fusion. Furthermore, an edge-aware detection head is designed to produce feature maps with clearer contour information, facilitating accurate target localization. By incorporating the rotation angle into the regression parameters and establishing a multi-parameter joint regression loss, our method achieves average accuracies of 94.23% and 81.10% on the SSDD+ and HRSID datasets, respectively, outperforming other SAR ship detection methods. The source code is available at <a href="https://github.com/LKE-HO/NSFANet">https://github.com/LKE-HO/NSFANet</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhanced rotating ship detection in SAR images via noise suppression and feature amplification

  • 期刊论文

  • 发表日期：

    2025年9月3日

查看全文
  • Kehong Liu,
  • Ming Zhang,
  • Yan Ren,
  • Yu Guo,
  • Guoqing Li,
  • Xiaoli Gao

摘要

Ship detection represents a significant application of synthetic aperture radar (SAR) technology. Despite its advantages, SAR-based ship detection models face challenges due to unique imaging characteristics, including substantial scattering noise and arbitrary ship orientations. To address these issues, we propose Enhanced rotating ship Detection in SAR Images via Noise Suppression and Feature Amplification (NSFA-Net) for rotating SAR ship image detection. NSFA-Net introduces a frequency domain attention module to highlight foreground targets and suppress background noise, enabling the construction of an effective feature extraction network. An optimized aggregation network-GR is employed to mitigate redundant noise generated during feature fusion. Furthermore, an edge-aware detection head is designed to produce feature maps with clearer contour information, facilitating accurate target localization. By incorporating the rotation angle into the regression parameters and establishing a multi-parameter joint regression loss, our method achieves average accuracies of 94.23% and 81.10% on the SSDD+ and HRSID datasets, respectively, outperforming other SAR ship detection methods. The source code is available at https://github.com/LKE-HO/NSFANet.