Steel Surface Defect Detection Based on Improved YOLOv5

Abstract

There are some problems in steel surface defect detection in industrial production, such as low detection rate and high missed detection rate. An improved YOLOv5 detector is proposed, which adds multi-scale attention mechanism in the feature pyramid structure, so that the model can pay more attention to the interesting region. In the process of down-sampling, the detailed information of the image is compressed, which may lead to the loss of accuracy. A Bottom-Up Fusion (BUF) module is used for down-sampling to obtain rich semantic information of deep network. The experimental results show that the improved YOLOv5 model mAP reaches 62.97%, which is 3.76 ^% higher than the baseline, and the detection accuracy of all types has been improved, and there is no big change in parameter quantity