Scholarlink

<p>In challenging underwater environments with limited visibility due to poor lighting and murky water, the ability to detect and identify targets is essential for various marine operations. We present an improved YOLO11 network, called YOLO-CE, to increase the accuracy of underwater target recognition under such circumstances. First, we propose a new convolutional module (CRAConv) that integrates coordinate attention (CA) and receptive field attention (RFA) into the backbone network. Secondly, we propose an edge spatial fusion module (ESFM) and integrate it into C3k2 to naturally form C3k2-ESFM, which learns different image features more deeply. Then, content-guided attention (CGA) is embedded into the feature pyramid network (FPN) to constrain the consistency of feature fusion across the backbone and neck. To improve the precision and stability of target localization, we do not use the traditional CIoU but Wise-IoU v3. On the UTDAC2020 and URPC2021 datasets, the YOLO-CE algorithm achieves mAP50 scores of 85% and 82.7%, respectively. Our approach performs excellently in enhancing small targets and general detection precision compared to mainstream object detection algorithms. The source code of our YOLO-CE will be made available at <a href="https://github.com/lanfafa/YOLO-CE">https://github.com/lanfafa/YOLO-CE</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

YOLO-CE: an underwater low-visibility environment target detection algorithm based on YOLO11

  • 期刊论文

  • 发表日期：

    2025年4月10日

查看全文
  • Ruolan Chen,
  • Huibo Zhou,
  • Hui Xie,
  • Bingyang Wang

摘要

In challenging underwater environments with limited visibility due to poor lighting and murky water, the ability to detect and identify targets is essential for various marine operations. We present an improved YOLO11 network, called YOLO-CE, to increase the accuracy of underwater target recognition under such circumstances. First, we propose a new convolutional module (CRAConv) that integrates coordinate attention (CA) and receptive field attention (RFA) into the backbone network. Secondly, we propose an edge spatial fusion module (ESFM) and integrate it into C3k2 to naturally form C3k2-ESFM, which learns different image features more deeply. Then, content-guided attention (CGA) is embedded into the feature pyramid network (FPN) to constrain the consistency of feature fusion across the backbone and neck. To improve the precision and stability of target localization, we do not use the traditional CIoU but Wise-IoU v3. On the UTDAC2020 and URPC2021 datasets, the YOLO-CE algorithm achieves mAP50 scores of 85% and 82.7%, respectively. Our approach performs excellently in enhancing small targets and general detection precision compared to mainstream object detection algorithms. The source code of our YOLO-CE will be made available at https://github.com/lanfafa/YOLO-CE.