Scholarlink

<p>Underwater target detection is a critical and rapidly evolving research area with significant applications in both military and civilian fields. Multi-Beam Forward-Looking Sonar (MFLS) operates reliably in low-visibility conditions and is one of the most widely used technologies for underwater detection. However, the complex and dynamic underwater environment, signal attenuation and distortion, along with the high cost of signal acquisition and transmission, make MFLS images detection one of the most challenging tasks in computer vision and image processing. To address these challenges, this paper proposes a target detection model based on MFLS, named AGD-YOLO. First, the ADNet attention mechanism is introduced to enhance performance by focusing on relevant features while suppressing unrelated noise. This denoising mechanism balances efficiency and further improves the model’s detection performance. Second, MPDIoU is adopted as the boundary regression loss, which considers the overlapping region, center point distance, and deviations in width and height, thereby enhancing the efficiency and accuracy of bounding box regression. Third, a new dataset based on MFLS is constructed to facilitate the detection of Unmanned Underwater Vehicle (UUV). Experimental results show that the proposed model improves <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\({mAP}_{0.5:0.95}\)</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mrow> <mi mathvariant="italic">mAP</mi> </mrow> <mrow> <mn>0.5</mn> <mo>:</mo> <mn>0.95</mn> </mrow> </msub> </math></EquationSource> </InlineEquation> by 2.6% compared to the second-best detection algorithm, significantly enhancing detection performance on MFLS images.</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

AGD-YOLO: a forward-looking sonar target detection method with attention-guided denoising convolutional neural network

  • 期刊论文

  • 发表日期：

    2025年3月6日

查看全文
  • Shuangjie Fu,
  • Han Pan,
  • Jianzhe Huang,
  • Xuyang Zhang,
  • Zhongliang Jing

摘要

Underwater target detection is a critical and rapidly evolving research area with significant applications in both military and civilian fields. Multi-Beam Forward-Looking Sonar (MFLS) operates reliably in low-visibility conditions and is one of the most widely used technologies for underwater detection. However, the complex and dynamic underwater environment, signal attenuation and distortion, along with the high cost of signal acquisition and transmission, make MFLS images detection one of the most challenging tasks in computer vision and image processing. To address these challenges, this paper proposes a target detection model based on MFLS, named AGD-YOLO. First, the ADNet attention mechanism is introduced to enhance performance by focusing on relevant features while suppressing unrelated noise. This denoising mechanism balances efficiency and further improves the model’s detection performance. Second, MPDIoU is adopted as the boundary regression loss, which considers the overlapping region, center point distance, and deviations in width and height, thereby enhancing the efficiency and accuracy of bounding box regression. Third, a new dataset based on MFLS is constructed to facilitate the detection of Unmanned Underwater Vehicle (UUV). Experimental results show that the proposed model improves \({mAP}_{0.5:0.95}\) mAP 0.5 : 0.95 by 2.6% compared to the second-best detection algorithm, significantly enhancing detection performance on MFLS images.