Weak Supervised PCB Carbon Fuel Defects Detection Based On Neural Network

Printed Circuit Board(PCB)is used in almost all electronic products.With the rapid development of electronics,communication,computer and other industries,PCB products are also increasingly moving towards high performance,complexity,and density.An efficient and accurate PCB defect detection system has become an urgent need for the PCB industry.Automated Optical Inspection(AOI)is based on machine vision,combines electronics,digital image processing,photoelectric inspection and computer technology,and has developed rapidly in the field of precision industrial device inspection.Compared with traditional manual inspection and electrical inspection,the AOI method does not cause any impact on the PCB,and has high detection accuracy and speed.Relevant research on PCB defect detection based on AOI is also actively carried out at home and abroad.This paper applies machine vision and deep learning technology to PCB defect detection.It mainly researches the carbon oil region defect detection on PCB.The main research work of the paper is as follows:(1)PCB matching algorithm design.Based on the matching of the Mark point edge model,design Mark point selection criteria and use the edge detection operator to model,search and match the edges.And calculate the global affine transformation matrix based on the matching pixel point pairs.Finally,the PCB test and standard maps are implemented matching of the whole image;(2)Relocate and extract the image skeleton for the carbon oil region according to the distance transformation of the binary image and the Snake model,and implement defect detection through subdivision and comparison,achieving a detection accuracy of 92.04%;(3)Aiming at the limitations of traditional methods,with the help of deep learning methods,combined with HRNet and Dense Net,a feature extraction network was designed,and a positioning hotspot map was obtained by calculating classification loss and class activation mapping.(4)A deep learning PCB carbon oil defect detection data set was constructed using a GAN,and an open change detection data set was introduced.With the help of transfer learning,PCB carbon oil defect detection and positioning under weak supervision was achieved,and achieved 99.1% and 98.7% accuracy rates in change detection and defect detection,respectively.