Scholarlink

<p>Ship detection in SAR images is essential for maritime surveillance but is hindered by challenges like varying ship sizes and complex backgrounds. To enhance YOLOv8n for this task, we introduce three key innovations: the Multi-Scale Dual Attention (MDSA) module for better multi-scale feature fusion, the Frequency-Aware Feature Fusion (FAFF) module to capture both spatial and frequency domain information, and the Wise IoU v3 (WIoU v3) loss function to adjust gradient contributions based on anchor box quality. Our improved YOLOv8n model outperforms the baseline on the SSDD dataset, achieving gains of 2.32% in Precision, 3.42% in Recall, 3.55% in mAP@0.5, and 5.32% in mAP@0.5:0.95. Similarly, on the LS-SSDD dataset, it achieves gains of 2.66% in Precision, 2.94% in Recall, 5.73% in mAP@0.5, and 1.12% in mAP@0.5:0.95.</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

MFF-YOLO: enhanced ship detection in noisy synthetic aperture radar images with dynamic loss adjustment

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年7月30日

查看全文
  • Yufeng Yang,
  • Yizhao He

摘要

Ship detection in SAR images is essential for maritime surveillance but is hindered by challenges like varying ship sizes and complex backgrounds. To enhance YOLOv8n for this task, we introduce three key innovations: the Multi-Scale Dual Attention (MDSA) module for better multi-scale feature fusion, the Frequency-Aware Feature Fusion (FAFF) module to capture both spatial and frequency domain information, and the Wise IoU v3 (WIoU v3) loss function to adjust gradient contributions based on anchor box quality. Our improved YOLOv8n model outperforms the baseline on the SSDD dataset, achieving gains of 2.32% in Precision, 3.42% in Recall, 3.55% in mAP@0.5, and 5.32% in mAP@0.5:0.95. Similarly, on the LS-SSDD dataset, it achieves gains of 2.66% in Precision, 2.94% in Recall, 5.73% in mAP@0.5, and 1.12% in mAP@0.5:0.95.