Research On Surface Defect Detection Method Of Copper Clad Laminate Based On Machine Vision

Copper clad laminate is the core component of electronic industrial products.In the production process,due to the limitation of production process and the influence of production environment,copper deficiency,wrinkles,black spots,bubbles and scratches are prone to appear on the surface of copper clad plate,which seriously affect the qualified rate and use of products.It is particularly critical to detect the Copper clad laminate after production and eliminate the defective products.At present,most manufacturers use manual detection,but it is vulnerable to the influence of detection experience and subjective factors.It is an inevitable trend to replace it by machine vision detection,and the design of defect detection algorithm is the key.Starting from the production and manufacturing process of copper clad laminate,this paper studies the efficient defect detection algorithm of copper clad laminate for the purpose of accuracy and rapidity of detection.Firstly,a detection method based on secondary feature selection is proposed to realize the accurate detection of copper deficiency in edge defects.Secondly,in view of the difficulty in extracting the feature defects of folds,black spots,bubbles and scratches by traditional machine vision detection methods,a detection method based on ResNet50 migration model is proposed.The specific research contents and innovations of this paper are as follows:(1)In view of the high false detection rate of edge defects of copper clad laminate detected by traditional machine vision technology,a detection method based on secondary feature selection is proposed.For the first time,template matching and image morphology algorithm were used to detect suspected defects on the surface of copper clad laminate.However,warping and bending deformation were prone to occur during the processing of copper clad laminate,which affected the accuracy of template matching and led to many misjudgment defects in the first test results.Then,the feature region of the suspected defect is extracted by using the feature selection box,and its shape feature attribute value is calculated.The suspected defect is screened again by the support vector machine optimized by the modified bee colony algorithm.The experimental results show that the false detection rate of copper clad laminate defect detection can be controlled within 3%by secondary feature selection.(2)In view of the problems that the traditional machine vision is difficult to detect several types of surface defects of copper clad laminate,and the amount of defect data is small when using deep learning detection,a detection method based on ResNet50 migration model is proposed.The pre-trained ResNet50 model removes the full connection layer as a feature extractor,and redesigns the full connection layer and classification layer to form a new migration model.At the same time,this paper studies the influence of main parameters on the recognition performance through experiments,and obtains the variation curve between the accuracy of training set and the loss value.The experimental results show that the recognition accuracy of defects can meet the industrial requirements.