Research And Implementation Of PCB Defect Detection Based On Deep Learning

With the development of national digital transformation,Printed Circuit Board(PCB),which is a key component of electronic products,is widely used in medical,automotive electronics,aerospace and many fields.Due to technical limitations,the surface of PCB will produce a variety of defects in the production process.Therefore,in order to avoid the influence of PCB defects on the performance of electronic products,it is necessary to detect the defects of PCB quickly and accurately.The traditional PCB defect detection method has many problems such as high cost,low accuracy,slow speed and missed detection.To address the limitations of traditional methods,this thesis presents a defect detection method for PCB based on deep learning and realizes its systematic application.In this thesis,YOLOX is applied to PCB defect detection and used data augmentation to solve the problem that the original PCB dataset is less,and improve the accuracy of defect detection.First,in order to prevent over-fitting caused by small dataset,the original PCB defect dataset is expanded by using data augmentation and Super Resolution Generative Adversary Networks.Meanwhile,the expanded dataset is used to train the network,and compare it with the major object detection algorithm.The experimental results show that the mean average precision of PCB defect detection based on Yolox-l algorithm is 91.79%,and it can make the detection precision and speed balanced,and the detection performance is better than contrast algorithm,which is suitable for PCB defect detection in actual production.In order to further enhance the performance of Yolox-l and improve the detection performance of small target defects and complex defects,this thesis uses the Efficient Channel Attention(ECANet)to improve the Yolox-l.On the one hand,this thesis uses ECANet to optimize the effective feature layer.On the other hand,the feature pyramid network is improved by ECANet.The experimental results show that the mean average precision of PCB defect detection based on Yolox-l algorithm reaches 93%,which is 1.21% higher than original algorithm,especially the improved Yolox-l algorithm can better detect small defects and complex defects.Finally,this thesis uses the lightweight Flask framework to design and implement the PCB defect detection system.In addition,the stability of the system is further verified by changing the brightness,rotating the PCB image and adding Gaussian noise to simulate the actual PCB image.In order to improve the detection performance of PCB defects,this thesis proposes PCB defect detection algorithm based on YOLOX network and improved Yolox-l network,respectively.Finally,the PCB defect detection system based on the improved Yolox-l algorithm is implemented through the Flask framework.