Research On PCB Defects Automatic Inspection And Classification Based On Multi-feature Fusion

With the development of image processing,pattern recognition,industrial automation and computer technology,the application of computer vision technology in the manufacturing industry has attracted more and more attention.Automatic Optical Inspection(AOI)technology applied in Printed Circuit Board(PCB)production has achieved remarkable results.China is major country of PCB manufacturing.Research of AOI technology create great value for PCB manufacturing industry of China,include improving the speed and accuracy of quality inspection.Except improving production efficiency and saving labor costs,classification of defects can recover materials and improve production by analyzing the types of defects.This thesis focuses on the research of PCB defect detection and classification,and analyze the structure of a set of automatic inspection system and detection process.First of all,contour matching improved by ant colony optimization for continuous domains is used to achieve image registration,and the defect is detected by the image difference method.According to analysis of the types and characteristics of PCB defects,a classification model is trained by multi-feature fusion.Finally,we use directed acyclic graph support vector machine(DAGSVM)to recognize the types of PCB defects.The specific research work of this article is as follows:(1)A PCB defect detection system is proposed,including hardware components and software modules.The function modules,software structure and working process of the automatic inspection system are described.(2)Reveal the extracting method and descriptor of PCB defect features by research of PCB production process and analysis of PCB defects feature.The defects described with bag of visual words consists of shape,texture and color.(3)Study the registration method of images and analyze the characteristics of different types of registration methods.Contour matching is used to achieve fast image registration.In order to improve the accuracy of image registration,ant colony algorithm for continuous domain is used to obtain an optimal solution.(4)The type of complex PCB defects is recognize by discriminative model.This model is obtained by trains directed acyclic graph support vector machine with few and uneven defects sample set.Based on the above research,combined with the actual characteristics of the enterprise,the registration algorithm and classification algorithm are tested in the experimental environment to verify its feasibility and effectiveness.