Circuit surface defect detection method based on DDAD model

By introducing CBAM module and HWD downsampling module in the DDAD model, integrating channel and spatial information, the problem of low accuracy of surface defect detection in the traditional method is solved, efficient and accurate unsupervised detection is achieved, and the method is verified. Effectiveness and practicality.

CN119313619AActive Publication Date: 2025-01-14DONGGUAN UNIV OF TECH
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
CN202411349241.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-26
Publication Date
2025-01-14
Estimated Expiration
2044-09-26

AI Technical Summary

Technical Problem

The traditional IC circuit surface defect detection method based on supervised learning is difficult and time-consuming to collect positive samples, and the unsupervised diffusion model DDAD algorithm is difficult to utilize spatial information and channel information during image reconstruction, resulting in low detection accuracy.

Method used

A circuit surface defect detection method based on DDAD model was designed. By introducing CBAM module and HWD downsampling module, the denoising network is improved, channel information and spatial information are integrated, and the maximum pooling downsampling is replaced to HWD downsampling to improve detection accuracy. The rate does not require a lot of labeling work.

Benefits of technology

The accuracy and efficiency of surface defect detection of IC circuits have been significantly improved, the speed and accuracy of detection have been improved, and the AUROC index has been significantly improved, verifying the effectiveness and practicality of the method.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119313619A_ABST
    Figure CN119313619A_ABST
Patent Text Reader

Abstract

The invention discloses a circuit surface defect detection method based on a DDAD model. The method comprises the following steps: acquiring to-be-detected circuit surface image data; and inputting to-be-detected circuit surface image data into the DDADCH defect detection model to obtain a circuit surface defect identification result. A denoising network of a DDAD model is designed and improved, a brand-new ResBlock module is provided, a CBAM module and an HWD down-sampling module are introduced into the denoising network of the original DDAD model, maximum pooling down-sampling in the network is completely replaced by the HWD down-sampling module, and the utilization rate of the network to space information is improved; meanwhile, the CBAM module is introduced into the ResBlock to fuse the channel information and the space information, so that the detection accuracy is improved, a large amount of labeling work is not needed, and the efficiency is further improved.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Defect detection method based on generative adversarial network and attention

    CN114943694A

  • Industrial product surface defect detection method and device based on weak supervision

    CN116721071A

  • Additive manufacturing process defect detection method and system based on improved YOLOv8 algorithm

    CN117392097A

  • Industrial appearance defect detection method and equipment based on photometric stereo method and diffusion model

    CN118429284A

  • Systems and methods for powder bed additive manufacturing anomaly detection

    US20220134435A1