Deep Learning Based Defect Inspection Of High Density Flexible Integrated Circuit Substrate

Flexible integrated circuit substrate(FICS)is an important part of integrated circuits,and it is widely used in various electronic products.The development direction of electronic equipment is to achieve more functions in smaller volume.Therefore,in order to save the space of electronic products,FICS has been developing towards high density.At present,the line width and line spacing of FICS have reached the micron level.If the defective products were brought into the next process,it will lead to huge losses,so FICS defect detection is a key link in the production process.At present,most manufacturers still use manual visual inspection.However,manual visual inspection will lead to a lot of waste of labor resources,and high-density lines will make manual visual inspection very difficult and inefficient.If the traditional digital image feature technology is used for automatic defect detection of FICS image,because it needs to traverse the image at the pixel level,there is a problem of slow speed.In recent years,deep learning technology has developed rapidly.This paper explores the defect detection of high-density flexible circuit board based on in-depth learning technology,but FICS defect samples are few,and some defects are difficult to detect.So this paper studies the small sample problem and hard example problem in defect detection.The main work of this paper is as follows:(1)A defect detection algorithm based on improved Faster R-CNN is proposed.Based on the idea and structure of Faster R-CNN algorithm,a method of feature extraction by two paths is proposed,and the structure of parallel spatial transformer network is added.Finally,the idea of hard example mining is introduced into the classification and regression network to improve the Faster R-CNN algorithm,which is applied to FICS defect detection to solve small sample problem and hard examples problem,and improve the accuracy rate of defect detection.(2)An improved defect detection algorithm based on YOLOv3 is proposed.On the basis of YOLOv3 algorithm and its structure,a densely connected structure is added,and feature fusion is embedded to enhance the ability of the whole network model to extract the key information of FICS defects and solve the small sample problem.Then the loss function used in YOLOv3 is improved and the idea of de-dimensionalization is introduced.Finally,the improved YOLOv3 algorithm is applied to FICS defect detection task.This paper provides a reference for defect detection technology based on deep learning in manufacturing process of high density flexible integrated circuit substrate.