期刊文献+

基于可变形卷积和注意力的带钢瑕疵识别方法

Strip defect recognition method based on deformable convolution and attention
下载PDF
导出
摘要 带钢生产中瑕疵检测的准确性对于其质量的保证有着至关重要的意义。本文针对带钢表面瑕疵检测中的瑕疵种类复杂、背景干扰严重、瑕疵样式间面积形态差异较大等问题,提出了一种基于ResNet50改进的结合多尺度变形卷积和注意力的带钢表面瑕疵分类识别方法。首先通过细化ResNet50中BottleNeck结构的卷积块为一组多尺度卷积以扩大感受野,然后引入可变形卷积代替组中的卷积核,使网络在训练中捕捉不同形态尺度的瑕疵特征。最后在网络中引入增强注意力模块,使网络可以关注到通道与空间之间的信息,将其联合起来从而关注到更重要的通道和空间位置。通过对比实验表明,本文提出的方法在带钢的表面瑕疵分类识别上精确度优于现有方法,可以应用于企业实际的带钢的工业生产中。 The accuracy of defect detection in steel strip production is of great significance for ensuring its quality.This paper proposes a steel strip surface defect classification and recognition method based on ResNet50 improved by combining multi-scale deformable convolution and Enhanced Attention to address the problems of complex defect types,severe background interference,and significant differences in area and shape of defect patterns.Firstly,the convolution blocks of the BottleNeck structure in ResNet50 are refined into a group of multi-scale convolutions to expand the receptive field.Then,the deformable convolution is introduced to replace the convolution kernel in the group,enabling the network to capture defect features of different shapes and scales during training.Finally,enhanced attention module is introduced into the network to enable it to focus on the information between channels and space,which are combined to focus on more important channels and spatial positions.Comparative experiments show that the proposed method has better accuracy in steel strip surface defect classification and recognition than existing methods,and can be applied to industrial production of steel strips in enterprises.
作者 万燕 齐浩天 姚砺 WAN Yan;QI Haotian;YAO Li(School of Computer Science and Technology,Donghua University,Shanghai 201620,China)
出处 《智能计算机与应用》 2024年第5期61-66,共6页 Intelligent Computer and Applications
关键词 带钢表面瑕疵识别 多尺度 可变形卷积 注意力机制 steel strip surface defect recognition multi-scale deformable convolution attention mechanism
  • 相关文献

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部