PCB Bare Board Surface Defect Detection System Based On Image Processing

With the advent of the era of Industry 4.0,the manufacturing industry has begun to enter the development route of intelligence,high-end and green.Printed circuit board(PCB)is the support of electronic components,and the automatic detection of its surface defects is one of the main directions of future manufacturing development.At present,Automated Optical Inspection(AOI)equipment is commonly used for automatic detection of PCB surface defects.AOI equipment has high efficiency and high reliability,but its high price makes many enterprises discouraged.Therefore,it is of great practical significance to study economical AOI equipment.The thesis takes single-layer and double-layer PCB bare boards as the research object,designs a set of PCB bare board surface defect detection system based on image processing,and completes the hardware selection and software design of the system.The system software mainly includes three parts,template extraction module,defect detection module and auxiliary module.The defect detection module is the core of the system software and the core of this thesis.Image preprocessing is the first step in defect detection work.Firstly,for the problem of low contrast of PCB images,three image enhancement methods such as histogram equalization,image sharpening and linear transformation are compared,and linear transformation is finally selected as the image enhancement algorithm in this thesis.Secondly,in order to obtain a complete PCB image,stitch the images,and select the SURF(Speed Up Robust Feature)algorithm as the matching algorithm for stitching through the statistical results of the quality evaluation indicators of the matching algorithm.According to the characteristics of the images collected by the system and the scale invariance of the SURF algorithm,an image stitching algorithm with higher realtime performance is proposed.After preprocessing,the PCB is inspected for defects using the reference method.The matching pairs obtained by image stitching realize the registration of template image and sample image,and through image processing steps such as image difference,OTSU double-threshold segmentation and open operation,the positioning of PCB defects is completed,and a PCB defect data set is established.Then,compared the two neural network models of Res Net and Mobile Net V3,adopted Res Net as the classification model in this thesis,and combined Res Net and SPP(Spatial pyramid pooling)structure to propose a neural network model with better classification performance,and its classification accuracy can reach 99.79%.Finally,the thesis designs and implements the graphical user interface of the system software.At the same time,the overall test of the system is carried out.The test results show that the PCB defect detection system designed in this thesis fully meets the proposed technical indicators.The maximum missed detection rate is 0.95%,the maximum false detection rate is 3.33%,the maximum detection time of a single PCB is 8.455 s,and the hardware cost is low,which meets the actual needs of China's market.