Lightweight Surface Defect Detection Algorithm Based on Improved YOLOv5

In response to the current issues of low detection accuracy and slow detection speed of detection algorithms in detecting surface defects on metal surfaces, a lightweight surface defect detection algorithm based on improved YOLOv5 is proposed. Firstly, a lightweight spatial attention module is introduced into the SPP module to extract effective information in the spatial dimension. Secondly, a new feature fusion network Amsf is designed to replace the original feature fusion network, which integrates feature information of different scales. Furthermore, the FAM attention mechanism is added to this network to enhance the detection capability for different types of targets. Finally, the SIoU loss function is used to replace the GIoU loss function to reduce boundary loss errors and improve the accuracy of target localization. Experimental results show that the improved model achieves an average detection accuracy of 81.97% on the NEU-DET dataset, which is 7.74% higher than YOLOv5. The model size and detection speed are comparable to YOLOv5. Effective improvement in detection accuracy is achieved while ensuring detection speed.