Supervision-based and non-supervision-based defect detection method, model establishment method and device

By combining supervised and unsupervised learning branches in the defect detection model, the problems of incomplete coverage and insufficient accuracy in existing defect detection technologies are solved, achieving high-precision, stable, and robust defect detection results.

CN120931630AActive Publication Date: 2025-11-11ZHEJIANG SHUANGYUAN TECH CO LTD
10 Cites 2 Cited by

Patent Information

Application Number
CN202511446114.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-10-11
Publication Date
2025-11-11
Estimated Expiration
2045-10-11

AI Technical Summary

Technical Problem

In the existing technology, supervised defect detection methods are difficult to cover all defect types when defect sample collection is difficult and defect types are highly random, resulting in a high risk of missed defects. Unsupervised methods are insufficient in terms of detection accuracy and inference speed, especially in terms of poor detection capability for minor defects.

Method used

We employ a supervised learning branch based on the YOLO architecture and an unsupervised learning branch based on the anomaly detection architecture. We extract features through a shared feature extraction module and use a freeze-thaw strategy during training. We combine labeled samples and positive samples for joint training to build a defect detection model. We optimize the model using loss functions such as classification loss, bounding box regression loss, alignment loss, and consistency loss.

Benefits of technology

It significantly improves the precision and accuracy of defect detection, increases the recognition rate of small defects and defects with inconspicuous features, reduces the feature library size and inference latency, and achieves collaborative perception of supervised and unsupervised tasks as well as the stability and robustness of the model.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention provides a supervised and unsupervised defect detection method, a model building method and a model building device, and the model building method comprises the steps: building an initial defect detection model which comprises a shared feature extraction module based on a YOLO backbone network, a supervised learning branch based on a YOLO architecture and an unsupervised learning branch based on an anomaly detection architecture; freezing an unsupervised learning branch, and training the shared feature extraction module and a supervised learning branch based on a labeled sample; after training of the shared feature extraction module and the supervised learning branches is completed, the shared feature extraction module and the supervised learning branches are frozen, unsupervised learning branches are unfrozen, and the unsupervised learning branches are trained based on positive samples; unfreezing the supervised learning branch, performing joint training on the shared feature extraction module, the supervised learning branch and the unsupervised learning branch by adopting the labeled sample and the positive sample, and obtaining a defect detection model after the training is completed; according to the method, the defect detection precision can be improved.
Need to check novelty before this filing date? Find Prior Art