A Research Method For Surface Defect Detection Of Printed Circuit Board Based On Deep Learning

PCB is a key component of electronic equipment.With the development of integrated circuit packaging technology,PCB surface wiring becomes more and more crowded,which puts forward more stringent requirements for PCB quality.The quality of PCB directly determines the safety performance of electronic products,the existence of defective PCB will lead to electronic products can not work normally,even cause major safety accidents,so the research of PCB surface defect detection technology becomes particularly important.Traditional methods can not be used for accurate detection,because,there are few features available for extraction of PCB surface defects.In order to improve the detection accuracy and meet the needs of industrial production,this paper conduct in-depth research on PCB surface defect detection technology based on deep learning,build PCB intelligent detection system,and realize accurate detection of PCB surface defects.The research work is as follows:(1)The resolution of PCB defect images was improved.In order to solve the problem that the resolution of PCB defect image decreased after magnification,PCB data set was processed by SRGAN network to improve the resolution of image.(2)A Mobilenet V3-YOLOv4 algorithm based on ECA attention mechanism was proposed.K-Means algorithm was used to cluster data sets to accelerate network training.Mobilenet V3 was used as the backbone network of YOLOv4 algorithm to make the network lightweight and improve the detection speed.ECA attention mechanism was added to enhance the feature extraction capability of the network to improve the detection accuracy.By comparing the improved algorithm with YOLOv4,YOLOv3,SSD and Faster R-CNN,the simulation results showed that the improved YOLOv4 algorithm had better detection effect than other algorithms.(3)A Ghost Net-YOLOv7 algorithm based on CA attention mechanism was proposed.Bisecting K-means algorithm was used to solve the problem of local convergence of K-means algorithm.Through lightweight network comparison test,Ghost Net was selected as the backbone network of YOLOv7 algorithm to reduce the number of network parameters.The CA attention mechanism and Inceptionv3 structure were added to the network to reduce the accuracy loss caused by Ghost Net.The feasibility of the improved scheme was proved by ablation experiments.By comparing the improved algorithm with YOLOv7,improved YOLOv4 and Faster R-CNN,the simulation results showed that the improved YOLOv7 algorithm solved the problem of low defect detection accuracy in the YOLOv4 algorithm,and had better detection effect compared with other algorithms.(4)PCB defect intelligent detection system was established.The Python programming language combined with Py Qt5 interface design tool and My SQL database was used to build an “intelligent detection system for PCB” to realize real-time detection of six kinds of defects on PCB surface and voice alarm for detected defects.