基于改进YOLOv5的带钢表面缺陷检测

Surface defect detection of strip steel based on improved YOLOv5

针对目前热轧带钢表面缺陷检测算法精度不高,检测速度慢的问题,提出了一种基于改进YOLOv5算法的网络模型。首先,引入Coordinate Attention提高模型特征提取能力;其次,对Neck结构进行改进,提出CA-BiFPN网络结构减少信息特征流失,实现多尺度信息表征;最后,使用EIOU Loss作为边框回归损失函数,提高定位精度,加快检测速度。实验结果表明,在NEU-DET数据集上相较于原YOLOv5算法平均精度均值(mAP)提高4.3%,召回率提高5.5%,精度提高2.2%,检测速度为111.1 fps,实现了识别精度与检测速度的良好均衡,具有一定的应用价值。

A network model based on improved YOLOv5 algorithm was proposed to solve the problems of low accuracy and slow detection speed of hot rolled strip surface defects.Firstly,Coordinate Attention is introduced to improve the feature extraction ability of the model.Secondly,the Neck structure is improved,and the CA-BiFPN network structure is proposed to reduce the information feature loss and realize the multi-scale information characterization.Finally,EIOU Loss is used as the frame regression loss function to improve the positioning accuracy and speed up the detection.The experimental results show that compared with the original YOLOv5 algorithm,the mean precision(mAP)of NEU-DET data set is increased by 4.3%,the Recall is increased by 5.5%,the Precision is increased by 2.2%,and the detection speed is 111.1fps,which achieves a good balance between the identification accuracy and detection speed,and has certain application value。