Scholarlink

<p>Target detection in synthetic aperture radar (SAR) images is crucial for SAR image interpretation. However, current deep learning-based SAR target recognition methods often rely solely on single-polarization information, limiting their ability to fully leverage polarimetric features and enhance model interpretability. This paper proposes DAP-Net, a dual-channel target recognition network that integrates attention mechanisms and polarimetric information to address this limitation. DAP-Net consists of a dual-channel feature extraction module with squeeze-and-excitation networks to separately process SAR images under vertical–horizontal and vertical–vertical polarization modes. Additionally, an improved C3 module incorporating depthwise separable convolutions and a bottleneck attention module is introduced to enhance feature extraction and lighten the network. To tackle the issue of imbalanced SAR datasets, an improved focal loss function is proposed to prioritize difficult samples during training. Experiments on both a simulated SAR image dataset (ES-SAR) and the OpenSARShip dataset demonstrate that DAP-Net achieves superior performance compared to existing methods, increasing the mean average precision by 1.39% on ES-SAR and 1.08% on OpenSARShip. The results validate the effectiveness and feasibility of the proposed approach in enhancing SAR target recognition. The source code is available at <a href="https://github.com/yyyy0101/DAP-Net.git">https://github.com/yyyy0101/DAP-Net.git</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

DAP-Net: enhancing SAR target recognition with dual-channel attention and polarimetric features

  • 期刊论文

  • 发表日期：

    2025年2月8日

查看全文
  • Fang Zhou,
  • Tingting Yang,
  • Liuyan Tan,
  • Xiaolong Xu,
  • Mengdao Xing

摘要

Target detection in synthetic aperture radar (SAR) images is crucial for SAR image interpretation. However, current deep learning-based SAR target recognition methods often rely solely on single-polarization information, limiting their ability to fully leverage polarimetric features and enhance model interpretability. This paper proposes DAP-Net, a dual-channel target recognition network that integrates attention mechanisms and polarimetric information to address this limitation. DAP-Net consists of a dual-channel feature extraction module with squeeze-and-excitation networks to separately process SAR images under vertical–horizontal and vertical–vertical polarization modes. Additionally, an improved C3 module incorporating depthwise separable convolutions and a bottleneck attention module is introduced to enhance feature extraction and lighten the network. To tackle the issue of imbalanced SAR datasets, an improved focal loss function is proposed to prioritize difficult samples during training. Experiments on both a simulated SAR image dataset (ES-SAR) and the OpenSARShip dataset demonstrate that DAP-Net achieves superior performance compared to existing methods, increasing the mean average precision by 1.39% on ES-SAR and 1.08% on OpenSARShip. The results validate the effectiveness and feasibility of the proposed approach in enhancing SAR target recognition. The source code is available at https://github.com/yyyy0101/DAP-Net.git.