Image Registration And Image Semantic Analysis Generation And Processing For PCB Surface Defect Detectio

Printed circuit board(PCB)is the carrier for the electrical connection of electronic components in integrated circuits.With the development of the integrated circuit industry,defect detection of the upstream raw material PCB is important to ensure the safety and performance of electronic products in various downstream applications.This project investigates a robust and high-precision PCB surface defect detection method to promote the automation and intelligent development of the surface defect detection of industrial products represented by integrated circuit PCBs,to effectively improve the efficiency and accuracy of defect detection,to enhance the quality of industrial products,and to ensure the safe and stable operation of electronic products in various fields as well as the development of intelligent manufacturing.This topic mainly includes three research contents: image alignment,sample generation and defect detection.Among them,image alignment provides a reference base for defect detection,and sample generation builds a PCB defect annotation dataset to provide a data base for defect detection,so as to achieve higher accuracy defect detection.1.A novel edge-guided energy-based image alignment method is implemented.This algorithm converts the edge information of the template into energy values by means of an edge-guided energy conversion module,and iteratively optimizes the energy loss function of the target contour point cloud by using the consistency of the target contour and the template contour point cloud to estimate the optimal geometric transformation parameters to achieve image alignment.This method has high alignment accuracy and high robustness,which solves the problem of alignment errors caused by defective features and other influencing features matching in defect detection tasks that are difficult to solve by traditional feature alignment methods.2.A defect label image generation method using semantic editing and generative adversarial networks is implemented.The research starts with a defect semantic editing method based on the PCB prior knowledge,which precisely controls the shape of defects using edge and semantic information,considering defect principles and other priors.This leads to the generation of a novel defect semantic map.Subsequently,a generative adversarial network-based image generation method is explored.By utilizing a small number of real defect samples,the method learns the shape,color,and texture distribution of PCB defects.Finally,a pre-trained generative model is employed to transform the defect semantic maps into real images,providing diverse and unknown defect label data.This approach addresses the scarcity of defect images and associated labeled samples in the PCB domain,while also obtaining labeled datasets for defect detection and semantic segmentation,thereby laying the foundation for improving detection accuracy.3.a defect detection method based on semantic analysis is implemented,which involves a in-depth analysis and summarization of image and defect generation principles at the fundamental level.The study focuses on corresponding detection methods for structural defects and texture defects.For structural defects,an edge-guided energybased image alignment and PCB defect detection method is developed.Building upon the energy-based image registration,the method measures the abnormality score of the aligned contour point cloud in the test image as an anomaly metric,based on the inconsistency between the test image and the template image.By adjusting the anomaly threshold,the method accurately locates the defects with strong interpretability.For texture defects,an unsupervised anomaly detection algorithm based on PCB priors and Fast MCD is investigated,enabling preliminary segmentation of anomaly regions by learning the distribution of positive and negative samples.Furthermore,a supervised deep learning-based semantic segmentation algorithm is employed to address false positive issues in PCB detection,achieving high-precision detection of texture defects based on practical defect standards.