Printing defect detection model and printing defect detection method

Abstract

The invention discloses a printing defect detection model and a printing defect detection method, and relates to the field of artificial intelligence, two multi-scale skeleton networks with the same structure are utilized, semantic feature maps of a reference image Iref and a to-be-detected image Idet are extracted at the same time, and according to a plurality of semantic feature maps of the reference image Iref and a plurality of semantic feature maps of the to-be-detected image Idet, a printing defect detection result is obtained. And calculating the similarity between the reference image Iref and the to-be-detected image Idet, and judging whether the to-be-detected image Idet has the printing defect according to the similarity, so that the defect existing in the printed matter can be effectively detected, the detection performance is improved, manual marking is not needed when the printing defect detection model is trained, the training cost is greatly reduced, and the printing defect detection efficiency is improved. During use, debugging by professionals is not needed, and usability is improved.

Description

technical field [0001] The invention relates to the field of artificial intelligence, in particular to a printing defect detection model and a printing defect detection method. Background technique [0002] Printing defect detection is a branch of the field of industrial defect detection. Similar to other industrial defect detection problems, the goal of printing defect detection is to automate the quality inspection of printed products through machine vision technology, so as to reduce labor costs and improve detection efficiency and accuracy. Effect. The special feature of printing defect detection is that the defects of printed products are content-dependent, that is, the types and characteristics of defects will change with the change of printing content. [0003] The existing printing defect detection technology usually adopts the idea of ​​comparing the image to be detected and the template image. In the prior art, based on whether deep learning technology is used as...

AI Technical Summary

Problems solved by technology

[0004] The invention patent with application publication number CN111709909A discloses a general-purpose printing defect detection method and its model based on deep learning. The defect of this scheme is that the output of the pixel classification module is the predicted value at the pixel level, and the cross-entropy function is used as a loss function

Therefore, in order to complete the training, it is necessary to obtain pixel-level manual labeling, that is, it is necessary to label whether each pixel of the original image is a defect, resulting in high training costs

[0005] The invention patent with application publication number CN110956630A discloses a planar printing defect detection method, device and system. The defects of this solution are: image preprocessing, image registration, and difference image calculation and analysis. buckle, there can only be a very small error, resulting in poor robustness, and it needs to be debugged by professionals when using it

Image registration, difference image calculation and other steps are all carried out on the original image. Due to the existence of lighting conditions, mechanical errors and other factors, it is often difficult to ensure the stability of the imaging quality, so it is easy to cause problems such as poor registration effect, which in turn affects the detection results.

When in use, this technology requires professionals to debug a large number of parameters according to the actual situation, which is time-consuming and laborious. Once the debugging results are not ideal, the detection performance will be greatly reduced, and the usability is poor.

Images