Defect detection model training method, device, electronic device and storage medium
By jointly training the encoder, classifier, domain adaptation module and decoder in the defect detection model, and using sample image labels to extract domain-invariant features, the problem of decreased accuracy in detecting new objects is solved, and the adaptability and accuracy of the detection model are improved.
CN114781605BActive Publication Date: 2025-10-03GUANGDONG UNIV OF TECH
3 Cites 0 Cited by
Patent Information
- Application Number
- CN202210396947.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-04-15
- Publication Date
- 2025-10-03
- Estimated Expiration
- 2042-04-15
AI Technical Summary
Technical Problem
The existing defect detection model has a decreased detection accuracy when faced with new items and is difficult to adapt to target items in different types of packaging.
Method used
By jointly training the encoder, classifier, domain adaptation module and decoder in the defect detection model, and utilizing the category labels, domain labels and segmentation labels of sample images, domain-invariant features are extracted, the impact of sample domain offset is reduced, and the detection accuracy is improved.
Benefits of technology
The accuracy of defect detection for new items in the target domain is improved, and the degradation of model performance caused by data distribution differences is reduced.
✦ Generated by Eureka AI based on patent content.
Abstract
The present application provides a defect detection model training method, device, electronic device and storage medium for improving the problem of gradually decreasing accuracy of defect detection using a model. The method includes: obtaining a sample image and a category label, domain label and segmentation label corresponding to the sample image, the category label characterizing whether the sample image has a defect, and the domain label characterizing whether the data domain to which the sample image belongs is a source domain sample or a target domain sample; using the sample image and the category label corresponding to the sample image to train the encoder and classifier in the defect detection model, and using the sample image to train the encoder and domain adaptation module in the defect detection model, and using the sample image and the segmentation label corresponding to the sample image to train the encoder and decoder in the defect detection model, obtaining a defect detection model by joint training, and the defect detection model is used to detect defects in images.
Need to check novelty before this filing date? Find Prior Art