Defect detection of printed circuit board based on improved YOLOv5

Abstract

Aiming at the problems of low efficiency, poor universality and high false detection rate in the existing PCB defect detection methods, a PCB defect detection algorithm P-YOLOv5 based on improved YOLOv5 is proposed. Firstly, the convolution block attention module is embedded in the backbone network to improve the feature extraction ability of the model; Secondly, a larger detection scale is added to the model network to expand the detection range of the model and improve the detection effect of PCB small target defects; Finally, the EIoU loss function is used to replace the GIoU loss function as the bounding box regression loss function to improve the detection accuracy of the model. The model training part uses the PCB-DATASET dataset published by Beijing University. The results show that the mean average precision of P-YOLOv5 algorithm can reach 99.03%, which is 2.52% higher than the original YOLOv5 algorithm. At the same time, the detection speed reaches 47.62 FPS, which can meet the requirements of PCB real-time detection.