Scholarlink

<p>With the rapid advancements in remote sensing techniques and deep learning, ship detection has become widely adopted in ocean monitoring. However, challenges such as background interference and multi-scale target features continue to hinder effective ship detection. To cope with these challenges, this paper proposes the multi-scale ship detection approach (MSDA) aimed to enable efficient multi-scale ship detection in remote sensing images with complex backgrounds. Firstly, this paper proposes the edge focus enhancement module (EFEM), which improves the model’s capability to capture ship details by enhancing edge features, thereby reducing background interference. Additionally, the proposed dilated residual aggregation network (DRAN) integrates the single-shot aggregation network and the dilation-wise residual module, significantly boosting the model’s capacity to learn and express multi-scale information. Finally, the paper employs the weighted bidirectional feature pyramid network (BiFPN) to optimize feature fusion, effectively accounting for the contributions of features across different resolutions, thereby further improving detection performance. Experimental results on the HRSC2016, Seaship7000, and SSDD datasets demonstrate that MSDA significantly improves performance, effectively addressing background interference and multi-scale target detection challenges. Furthermore, this paper validates the inference speed of the method in a high-performance computing environment, demonstrating its potential to support real-time ocean monitoring applications.</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

A robust multi-scale ship detection approach leveraging edge focus enhancement and dilated residual aggregation

  • 期刊论文

  • 发表日期：

    2025年11月6日

查看全文
  • Datao You,
  • Bingbing Zhao,
  • Deyu Lei,
  • Yanxu Mao

摘要

With the rapid advancements in remote sensing techniques and deep learning, ship detection has become widely adopted in ocean monitoring. However, challenges such as background interference and multi-scale target features continue to hinder effective ship detection. To cope with these challenges, this paper proposes the multi-scale ship detection approach (MSDA) aimed to enable efficient multi-scale ship detection in remote sensing images with complex backgrounds. Firstly, this paper proposes the edge focus enhancement module (EFEM), which improves the model’s capability to capture ship details by enhancing edge features, thereby reducing background interference. Additionally, the proposed dilated residual aggregation network (DRAN) integrates the single-shot aggregation network and the dilation-wise residual module, significantly boosting the model’s capacity to learn and express multi-scale information. Finally, the paper employs the weighted bidirectional feature pyramid network (BiFPN) to optimize feature fusion, effectively accounting for the contributions of features across different resolutions, thereby further improving detection performance. Experimental results on the HRSC2016, Seaship7000, and SSDD datasets demonstrate that MSDA significantly improves performance, effectively addressing background interference and multi-scale target detection challenges. Furthermore, this paper validates the inference speed of the method in a high-performance computing environment, demonstrating its potential to support real-time ocean monitoring applications.