Scholarlink

<p>Sonar target detection is crucial for ensuring maritime safety and supporting resource exploration. However, the complex and dynamic underwater environment presents significant challenges in accurately detecting multiscale targets. To balance detection performance and accuracy, we propose a novel sonar target detector based on You Only Look Once version 10 (YOLOv10). Our approach utilizes a redesigned backbone network that integrates Space-to-Depth Convolution (SPD-Conv) and a Mixed Local Channel Attention (MLCA) module to enhance detection performance in complex underwater environments. We also introduce the Content-Guided Feature Pyramid Network (CFPN), which combines a pyramid structure and the Content-Guided Attention Fusion (CGA-Fusion) module to enhance feature fusion and improve detection of small underwater targets. Extensive experiments on publicly available forward-looking sonar image datasets show that our model achieves a 2.4% improvement in mean Average Precision (mAP), a 1.6% increase in precision and a 3% boost in recall rate. Additionally, the inference time remains largely unchanged.</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

SMC-YOLO: efficient object detector for underwater small sonar target

  • 期刊论文

  • 发表日期：

    2025年7月8日

查看全文
  • Bingru Li,
  • Runze Zhang,
  • Xudong Xu

摘要

Sonar target detection is crucial for ensuring maritime safety and supporting resource exploration. However, the complex and dynamic underwater environment presents significant challenges in accurately detecting multiscale targets. To balance detection performance and accuracy, we propose a novel sonar target detector based on You Only Look Once version 10 (YOLOv10). Our approach utilizes a redesigned backbone network that integrates Space-to-Depth Convolution (SPD-Conv) and a Mixed Local Channel Attention (MLCA) module to enhance detection performance in complex underwater environments. We also introduce the Content-Guided Feature Pyramid Network (CFPN), which combines a pyramid structure and the Content-Guided Attention Fusion (CGA-Fusion) module to enhance feature fusion and improve detection of small underwater targets. Extensive experiments on publicly available forward-looking sonar image datasets show that our model achieves a 2.4% improvement in mean Average Precision (mAP), a 1.6% increase in precision and a 3% boost in recall rate. Additionally, the inference time remains largely unchanged.