基于改进 YOLOv11n的轴承表面缺陷检测方法
CSTR:
作者:
作者单位:

(1.华北理工大学电气工程学院,河北唐山 063210; 2.河北省风光氢储安全监测与智能运行技术创新中心,河北唐山 063210)

作者简介:

周建新(1977—),男,河北省唐山市人,副教授,硕士生导师,主要研究方向为智能控制理论及应用;焦江涛(2002—),男,河北省邯郸市人,硕士研究生,研究方向为智能控制与模式识别。

通讯作者:

中图分类号:

TP391. 41;TH133. 3

基金项目:

河北省自然科学基金项目(F2018209201).


Bearing Surface Defect Detection Method Based on Improved YOLOv11n
Author:
Affiliation:

(1. College of Electrical Engineering,North China University of Science and Technology,Tangshan 063210,CHN;2. Hebei Wind-Solar Hydrogen-Storage System Security Monitoring and Intelligent Operation Center of Technology Innovation,Tangshan 063210,CHN)

Fund Project:

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    针对轴承表面缺陷检测中存在的检测精度不足和模型效率不高等问题,文章提出一种改进 YOLOv11n的缺陷检测模型 GSP-YOLOv11n。首先,引入级联组注意力机制 CGA,增强多头自注意力的上下文建模能力,提高对复杂纹理缺陷的感知精度。其次,引入 Slim-Neck轻量化特征融合结构,在保证多尺度特征有效传递的同时,显著降低了计算复杂度。最后,引入一种分裂式空间压缩卷积结构 SPConv,通过压缩冗余特征图,不仅有效减少了模型的计算量,还减少了参数规模,从而整体提升了模型的计算效率与轻量化程度。在自建轴承缺陷数据集上的实验结果表明,与原始 YOLOv11n相比,GSP-YOLOv11n的 mAP@0. 5和 mAP@0. 5∶0. 95分别提升了 2. 9%和 2. 0%,参数量减少约 9. 7%,计算量降低 1.2 G。该模型在轴承缺陷检测任务中表现优异,尤其对小目标和复杂形状缺陷的检测具有显著优势。

    Abstract:

    To mitigate drawbacks such as insufficient detection accuracy and low model efficiency in bearing surface defect detection,a defect detection model,GSP-YOLOv11n,based on improved YOLOv11n is proposed. First,the cascade group attention mechanism is introduced to enhance the context modeling ability of multihead self-attention and improve the perception accuracy of complex texture defects. Second,the Slim-Neck lightweight feature fusion structure is incorporated, which ensures the effective transfer of multiscale features while significantly reducing the computational complexity. Finally, a split spatial compression convolution structure SPConv is introduced. Compressing redundant feature maps effectively reduces the computational cost of the model and the parameter scale,thus improving the overall computational efficiency and lightweight degree of the model. Experiments on a self-built bearing defect dataset demonstrate that GSP-YOLOv11n outperforms the original YOLOv11n model in terms of mAP@0. 5(2. 9% enhancement),mAP@0. 5∶ 0. 95(2. 0% enhancement),the number of parameters(9. 7% reduction),and the computational cost(1. 2 G reduction). Thus,the proposed model shows high bearing-defect-detection performance,offering significant advantages for the detection of small targets and defects with complex shapes.

    参考文献
    相似文献
    引证文献
引用本文

周建新,焦江涛,王彬.基于改进 YOLOv11n的轴承表面缺陷检测方法[J].半导体光电,2026,47(2):380-387. ZHOU Jianxin, JIAO Jiangtao, WANG Bin. Bearing Surface Defect Detection Method Based on Improved YOLOv11n[J].,2026,47(2):380-387.

复制
分享
相关视频

文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2025-10-02
  • 最后修改日期:
  • 录用日期:
  • 在线发布日期: 2026-05-25
  • 出版日期:
文章二维码