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

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

作者简介:

通讯作者:

中图分类号:

基金项目:

河北省自然科学基金(F2018209201)


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

School of Electrical Engineering, North China University of Science and Technology

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,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,引入了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,同时减少了参数规模,从而整体提升了模型的计算效率与轻量化程度。通过在自建轴承缺陷数据集上进行实验,GSP-YOLOv11n在mAP@0.5和mAP@0.5:0.95方面相比原YOLOv11n模型分别提高了2.9%、2.0%,且相比于原YOLOv11n模型参数量减少约9.7%,计算量降低了1.2G。该模型在轴承缺陷检测任务中具有较好的性能,尤其对小目标和复杂形状的缺陷检测具有显著优势。

    Abstract:

    In view of the problems 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. Firstly, the cascade group attention mechanism CGA is introduced to enhance the context modeling ability of multi-head self-attention and improve the perception accuracy of complex texture defects. Secondly, the Slim-Neck lightweight feature fusion structure is introduced. While ensuring the effective transfer of multi-scale features, it significantly reduces the computational complexity. Finally, a split spatial compression convolution structure SPConv is introduced. By compressing redundant feature maps, it not only effectively reduces the computational amount of the model, but also reduces the parameter scale, thus overall improving the computational efficiency and lightweight degree of the model. Through experiments on the self-built bearing defect dataset, compared with the original YOLOv11n model, GSP-YOLOv11n has increased by 2.9% and 2.0% in mAP@0.5 and mAP@0.5:0.95 respectively, and the number of parameters has decreased by about 9.7% compared with the original YOLOv11n model, and the computational amount has decreased by 1.2G. This model has good performance in the bearing defect detection task, especially has significant advantages in 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
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:2025-11-17
  • 录用日期:2025-11-25
  • 在线发布日期:
  • 出版日期:
文章二维码