摘要
针对目前带钢表面缺陷检测方法无法适应缺陷形状不规则、尺度不一、背景复杂等问题,提出了一种基于Faster-RCNN的改进网络,在主干网络中采用可变形卷积模块,FPN多尺度检测模块及在RPN网络中融合CBAM注意力模块。实验结果表明,改进后的网络可有效地提高带钢表面缺陷检测的精度。
With regard to the problem that the existent steel strip surface defect detection method was incapable of adapting to these conditions such as irregular shapes,different scales,and complex backgrounds of surface defects,the improved network based on the Faster-RCNN Algorithm was proposed.Depending on the network,the deformable convolution modules and FPN(Feature Pyramid Networks)multiple dimensioned detection modules were used in the core network,and also CBAM(Convolutional Block Attention Module)attention modules were syncretized in the RPN(Region Proposal Network).The experimental results showed that the improved network could effectively improve the degree of accuracy of testing surface defects steel strips.
作者
吴健生
王健全
付美霞
王振乾
卢一凡
WU Jiansheng;WANG Jianquan;FU Meixia;WANG Zhenqian;LU Yifan(University of Science and Technology Beijing,Beijing 100083,China)
出处
《鞍钢技术》
CAS
2022年第6期23-28,32,共7页
Angang Technology
基金
国家重点研发计划(2020YFB1708800)
广东省重点研究与开发计划(2020B010113007)
广东省基础与应用基础研究基金联合基金(2021A1515110577)
中央高校基础研究基金项目(FRF-MP-20-37)
北京科技大学青年教师学科交叉研究项目(中央高校基本科研业务费专项资金)资助项目(FRF-IDRY-21-005)
中国博士后科学基金(2021M700385)的支持。
