A defect detection method based on dictionary learning

Abstract

The invention discloses a defect detection method based on dictionary learning, and the method comprises the following steps: collecting a standard panel image, extracting an edge information featureof the panel image, and enabling the edge information feature to become a feature image 1; Superposing and dividing the first feature image into small blocks with the same size to form a feature imagetraining library; Carrying out dictionary learning on small blocks in the feature image training library to obtain a feature image dictionary and a corresponding representation coefficient matrix; Collecting a to-be-detected panel image, extracting edge information characteristics of the panel image, and converting the edge information characteristics into a characteristic image II; Overlapping and dividing the feature image II into small blocks with the same size; Performing sparse representation on the small blocks obtained by each feature image II under the feature image dictionary to obtain corresponding sparse representation coefficients; Wherein if the L1 norm of the sparse representation coefficient is far greater than zero, it is considered that no defect exists on the small block, and if the L1 norm of the sparse representation coefficient is close to zero, it is considered that the defect exists on the small block, and then the to-be-detected panel has the defect.

Description

technical field [0001] The invention relates to the technical field of defect detection, in particular to a defect detection method based on dictionary learning. Background technique [0002] With the continuous improvement of people's living standards and the continuous popularization of electronic equipment, the production and shipment of electronic equipment is very large. How to realize the automatic detection of defects in the production process of electronic equipment products has become one of the concerns of major manufacturers. [0003] Among them, there are many steps in the production process of the glass cover plate of electronic equipment, there are many problems, and there is a lot of room for improvement. The main problems are: [0004] At present, the commonly used printing defect detection methods usually use the edge extraction algorithm based on the glass cover image, and then analyze the contour and the flatness and concave-convex area of ​​the edge, but ...

AI Technical Summary

Problems solved by technology

[0004] At present, the commonly used printing defect detection methods usually use the edge extraction algorithm based on the glass cover image, and then analyze the contour and the flatness and concave-convex area of ​​the edge, but the existing technology is still unable to detect subtle defects, because in the subtle Near the defect, the commonly used edge extraction algorithm cannot accurately extract the edge

[0005] In addition, the existing process uses manual visual inspection during the final inspection, which cannot effectively prevent damage to the glass cover caused by machine failure or damage. From the first occurrence of damage to manual detection during the final inspection, a series of The damaged parts cannot be adjusted in time, resulting in loss and waste of the manufacturer

Images