Research On PCB Defect Detection Based On Deep Learning

Printed circuit board(PCB)is an important part of the connection between mechanical and electronic components,with good market prospects.Its quality affects many industries,and rapid and accurate detection of PCB defects is of great research significance.Through research on the current research status of PCB defect detection methods both domestically and internationally,it was found that deep learning based PCB defect detection has become a research hotspot.However,this detection method has the problem of low accuracy in detecting small target defects in PCBs.Furthermore,conducting research on deep learning algorithms for PCB defect detection,it was found that YOLOv5 s has a high defect detection rate but low accuracy for PCBs.In order to achieve high PCB defect detection performance and speed,and meet PCB detection needs,this article conducts improvement research on it.The main content includes:1.The improvement research is carried out on the backbone network.The improvement method mainly increases the feature information of the high-level network about the shallow network by strengthening the connection between the high-level and shallow layers of the backbone network,so that the final output of YOLOv5 s network contains more feature information about PCB defects.The experiment shows that the improved method can improve the PCB defect detection accuracy of YOLOv5s;2.Yolo Head uses a single convolution operation,making it difficult to focus on extracting features.Therefore,a self attention mechanism is proposed.Furthermore,in order to address the issue of reducing the detection rate of YOLOv5 s by introducing a self attention mechanism,this article draws on the computational characteristics of SENet,CBAM,and residual units to optimize the self attention mechanism.The experiment shows that the introduction of optimized self attention mechanism in the Yolo Head layer runs faster than the introduction of original self attention mechanism,and the improvement of PCB defect detection accuracy in YOLOv5 s is similar to that of original self attention.3.Combining backbone improvement methods with optimized self attention mechanism to improve YOLOv5 s,and conducting experimental research from two perspectives: introducing single optimized self attention and introducing double optimized self attention.The experimental results show that the introduction of dual optimization self attention has the best improvement effect.The improved YOLOv5 s algorithm can detect PCB defect datasets with a maximum detection accuracy of96.14%,which is 4.68% higher than YOLOv5 s.The detection rate is 39.7 FPS and can be quickly detected.