Scholarlink

<p>Ship detection in synthetic aperture radar (SAR) imagery faces challenges due to complex maritime environments and varying ship sizes. Traditional deep learning methods often struggle with accuracy and computational efficiency, especially for small targets. This paper introduces ECM-YOLO, a detection framework based on YOLOv11, incorporating three key innovations: a multi-scale edge information fusion module (MSEIF) for enhanced feature extraction, a multi-scale attentional fusion module (MSAF) for improved small vessel detection, and a Shared Detail-enhanced Convolutional Detection Head (SDCD) for reduced complexity. Testing on HRSID and SSDD datasets showed ECM-YOLO achieves 93.3% and 98.0% mAP50, respectively, with an 8.5% parameter reduction, demonstrating superior accuracy and efficiency. The code is available at <a href="https://github.com/fciasth/EMO-YOLO">https://github.com/fciasth/EMO-YOLO</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

Multi-scale edge information fusion with 3D attention for ship detection in complex SAR imagery

  • 期刊论文

  • 发表日期：

    2026年5月28日

查看全文
  • Xinghai Zhao,
  • Pengfei Liu,
  • Chongxin Fang,
  • Junchen Liu,
  • Bo Yan

摘要

Ship detection in synthetic aperture radar (SAR) imagery faces challenges due to complex maritime environments and varying ship sizes. Traditional deep learning methods often struggle with accuracy and computational efficiency, especially for small targets. This paper introduces ECM-YOLO, a detection framework based on YOLOv11, incorporating three key innovations: a multi-scale edge information fusion module (MSEIF) for enhanced feature extraction, a multi-scale attentional fusion module (MSAF) for improved small vessel detection, and a Shared Detail-enhanced Convolutional Detection Head (SDCD) for reduced complexity. Testing on HRSID and SSDD datasets showed ECM-YOLO achieves 93.3% and 98.0% mAP50, respectively, with an 8.5% parameter reduction, demonstrating superior accuracy and efficiency. The code is available at https://github.com/fciasth/EMO-YOLO.