Research On PCB Defect Detection Algorithm Based On Deep Learning

With the rapid development of printed circuit boards(PCB)downstream industries such as computers,mobile phones,and smart vehicles,electronic products such as mobile phones and computers are ubiquitous.As PCB is an essential component,the presence of defects directly affects the extent to which an item meets quality standards.Detecting flaws in PCBs is therefore of the utmost importance.This study discusses the PCB defect detection method based on deep learning to address the issue that classical defect detection methods are cumbersome and time-consuming for use and cannot adapt to advanced industrial production efficiency.The primary contents of the research are as follows:(1)A PCB defect detection technique based on multi-scale feature fusion is proposed in light of the YOLOv5 algorithm’s poor detection effect on small target defects.The algorithm first adds a shallower feature fusion path to the three feature fusion paths in the feature fusion part and adds a detection head to identify this path because there are several little target defect information remnants in the shallow feature map.predict.In order to lessen the issue of information loss brought on by the network’s convolution operation,the feature extraction network is down-sampled to 8 times and 16 times the feature layer and then directly spliced with the output feature layer of the same size.Finally,the original network’s CIo U Loss is replaced by SIo U Loss.The angle of the prediction frame is limited to enhance the network’s detection efficiency by increasing the angle penalty index.It is proven that the optimization of the suggested approach is successful in enhancing the performance of the original network model by setting up ablation and comparison experiments.The final m AP value is 98.4%,which is higher than some of the most widely used detection methods currently in use.This indicates that the enhanced algorithm is successful in detecting PCB defects.field offers some benefits.(2)For the complex background of the PCB image and the variety of defect forms that are likely to lead to network misdetection,a multi-scale feature fusion technique built on the attention mechanism is provided.Based on the multi-scale feature fusion detection technique’s network structure,the algorithm initially substitutes the SPD module for the original algorithm’s feature extraction network’s downsampling operation to minimise information loss.This is done to minimize information loss during the network downsampling process.Additionally,the Swin-Transformer coding unit is employed to enhance the feature.The utilization of an extraction network can enhance the network’s capacity to acquire target defect information by capturing global and rich contextual information.Additionally,the Sim AM module is implemented to allocate three-dimensional attention weights to feature maps,ultimately improving the precision of target defect detection.The experimental findings indicate that the suggested method can reduce the misdetection of defective targets.The final m AP value reaches 98.8%,demonstrating the usefulness of the modified approach.