Scholarlink

<p>Ship object detection is a pivotal task in intelligent maritime sensing, vital for applications such as maritime surveillance, national defence, and traffic management. However, this task faces challenges due to variations in pixel occupancy across multi-scale objects, uneven object distributions in complex environments, and the presence of fine-grained ship categories with small inter-class differences and large intra-class variations. To address these issues, we propose enhanced multi-scale and fine-grained object detection with content-aware refinement, a novel network for multi-scale and fine-grained ship detection, called MFNet. MFNet incorporates a Content-aware Fusion Network and a Feature Refine Module. The Content-aware Fusion Network efficiently fuses multi-scale features and learns contextual information, enhancing representational ability of muti-scale features. The Feature Refine Module captures discriminative fine-grained features, improving detection accuracy. Extensive experiments on challenging datasets, including VOC, Seaships, HRSID, and MIB-Ships, demonstrate the effectiveness of MFNet, achieving state-of-the-art performance on the MIB-Ships dataset. Our code is available at: <a href="https://github.com/Wu-Junbao/MFNet">https://github.com/Wu-Junbao/MFNet</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 multi-scale and fine-grained object detection with content-aware refinement

  • 期刊论文

  • 发表日期：

    2026年5月20日

查看全文
  • Junbao Wu,
  • Hao Meng,
  • Ming Yuan,
  • Shouwen Cai

摘要

Ship object detection is a pivotal task in intelligent maritime sensing, vital for applications such as maritime surveillance, national defence, and traffic management. However, this task faces challenges due to variations in pixel occupancy across multi-scale objects, uneven object distributions in complex environments, and the presence of fine-grained ship categories with small inter-class differences and large intra-class variations. To address these issues, we propose enhanced multi-scale and fine-grained object detection with content-aware refinement, a novel network for multi-scale and fine-grained ship detection, called MFNet. MFNet incorporates a Content-aware Fusion Network and a Feature Refine Module. The Content-aware Fusion Network efficiently fuses multi-scale features and learns contextual information, enhancing representational ability of muti-scale features. The Feature Refine Module captures discriminative fine-grained features, improving detection accuracy. Extensive experiments on challenging datasets, including VOC, Seaships, HRSID, and MIB-Ships, demonstrate the effectiveness of MFNet, achieving state-of-the-art performance on the MIB-Ships dataset. Our code is available at: https://github.com/Wu-Junbao/MFNet.