Film defect detection method, device, apparatus, product, and storage medium
By generating diverse defect samples using a generator and training the model using a meta-learning strategy, the problems of sample scarcity and texture variation in membrane material defect detection are solved, achieving high-precision and fast-adaptive detection results while reducing data preparation costs.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHENZHEN INSTITUTE OF INFORMATION TECHNOLOGY
- Filing Date
- 2026-04-22
- Publication Date
- 2026-07-21
AI Technical Summary
In membrane material defect detection, existing technologies rely on a large amount of high-quality labeled data, especially since rare defect samples are scarce, resulting in low detection accuracy and recall. Furthermore, the models struggle to adapt quickly to texture changes and novel defects, increasing deployment and maintenance costs.
A diverse range of defect samples are generated by a generator, and a pre-defined membrane defect detection model is trained using a meta-learning strategy. An attention mechanism is used to decouple defect and background features, and data augmentation and loss function optimization are performed to construct a high-precision membrane defect detection model.
Achieving high-precision membrane defect detection under limited sample conditions reduces reliance on real samples, improves the model's generalization ability and adaptability, and lowers data acquisition and annotation costs.
Smart Images

Figure CN122434856A_ABST