Multi-scale defect detection of printed circuit board based on feature pyramid network

Printed circuit board (PCB) is one of the most important components in electronic products, and with the improvement of production technology, the structure PCB is more and more refined, so it is very important for the quality inspection of PCB. The traditional PCB defect detection methods are slow, error prone and the cost of detection is high. With the development of deep learning, there are many excellent object detection models, in this paper, the object detection model feature pyramid network is applied to PCB defect detection, in the network, PCB defects of different sizes are detected by constructing feature maps of different scales. In order to verify the effectiveness of the defect detection method, experiments are carried out on two bare PCB datasets and one PCB dataset with devices, and F1-score and ROC curve were used to evaluate the experiments, which proves the effectiveness of the defect detection network.