改进YOLOv5s的钢材表面缺陷检测算法

Improved YOLOv5 Algorithm for Steel Surface Defect Detection

为提高钢材表面缺陷检测的准确率,提出一种改进YOLOv5s的钢材表面缺陷检测算法。首先,在特征提取网络中引入Swin Transformer结构,增强网络对特征的感知能力;其次,添加坐标注意力机制,加强对重要特征信息的关注;最后,针对钢材缺陷的特点增加检测层,提升多尺度目标检测能力,并使用SIOU损失函数评估检测效果。将所提出的算法在公开数据集NEU-DET上进行消融实验,结果表明:所提算法能有效提高钢材表面缺陷目标检测的准确率。

For purpose of improving the accuracy in detecting steel surface defects,an improved YOLOv5s algorithm for the steel surface defect detection was proposed.Firstly,having Swin Transformer structure introduced into the feature extraction network to enhance feature perception ability of the network;secondly,having coordinate attention mechanism employed to enhance the attention to important feature information;finally,through considering the characteristics of steel defects,having detection layer added to improve the detection ability of multi-scale target,and having SIOU lossfunction used to evaluate the detection effect.Im plementing ablation experiments of the proposed algorithm on the public data set NEU-DET shows that,the proposed algorithm can effectively improve the average accuracy in detecting steel surface's target defect.