<p>With the rapid development of the ocean economy and the continuous in-depth development of ocean resources, object detection technology is becoming increasingly important in fields such as ocean scientific research, ecological protection, fisheries management, and ocean engineering. However, the complexity of the underwater background environment, overlap between organisms, small objects have difficulty being detected, and practical application issues associated with large models all increase the difficulty of detection. An HCL-YOLO detection algorithm based on an improved YOLOv10n is proposed to solve the above problem. First, the localization enhancement module (LE) was proposed to solve the problem of covered objects. By extracting positional information in the space and channel, the model’s focus on global information is improved, thereby enhancing the model’s reasoning ability for covered objects. Second, the cross-scale feature fusion (CSFF) neck structure is proposed to enhance the model’s feature interaction ability and multi-scale object feature fusion ability, thereby improving the detection accuracy of small objects. Finally, HetConv was introduced and the C2f module was redesigned so that the backbone network takes the form of a heterogeneous kernel. This ensures the backbone’s feature extraction capabilities and reducing the model parameters and computational requirements. The proposed model was compared with multiple sets of experiments on the URPC, DUO, and ROUD datasets. The HCL-YOLO model has 40.7% fewer parameters and 27.4% fewer floating-point calculations than YOLOv10n, with a 24 increase in FPS. The <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11554_2025_1674_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="53" /> </InlineMediaObject> <EquationSource Format="TEX">\(\hbox {mAP}_{0.5}\)</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mtext>mAP</mtext> <mrow> <mn>0.5</mn> </mrow> </msub> </math></EquationSource> </InlineEquation> reached 83.9%, 86.6%, and 84.3% respectively. The experimental results show that the HCL-YOLO model performs excellently in underwater detection tasks.</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

Hcl-yolo: a lightweight and efficient underwater object detection algorithm

  • Xiuman Liang,
  • Teng Zhang,
  • Haifeng Yu,
  • Zhendong Liu

摘要

With the rapid development of the ocean economy and the continuous in-depth development of ocean resources, object detection technology is becoming increasingly important in fields such as ocean scientific research, ecological protection, fisheries management, and ocean engineering. However, the complexity of the underwater background environment, overlap between organisms, small objects have difficulty being detected, and practical application issues associated with large models all increase the difficulty of detection. An HCL-YOLO detection algorithm based on an improved YOLOv10n is proposed to solve the above problem. First, the localization enhancement module (LE) was proposed to solve the problem of covered objects. By extracting positional information in the space and channel, the model’s focus on global information is improved, thereby enhancing the model’s reasoning ability for covered objects. Second, the cross-scale feature fusion (CSFF) neck structure is proposed to enhance the model’s feature interaction ability and multi-scale object feature fusion ability, thereby improving the detection accuracy of small objects. Finally, HetConv was introduced and the C2f module was redesigned so that the backbone network takes the form of a heterogeneous kernel. This ensures the backbone’s feature extraction capabilities and reducing the model parameters and computational requirements. The proposed model was compared with multiple sets of experiments on the URPC, DUO, and ROUD datasets. The HCL-YOLO model has 40.7% fewer parameters and 27.4% fewer floating-point calculations than YOLOv10n, with a 24 increase in FPS. The \(\hbox {mAP}_{0.5}\) mAP 0.5 reached 83.9%, 86.6%, and 84.3% respectively. The experimental results show that the HCL-YOLO model performs excellently in underwater detection tasks.