A method for attention residual guided chip defect detection
By using a multi-granularity attention residual module and a dynamic parameter update method, the problems of feature overwhelming and noise accumulation in laser chip defect detection are solved, improving detection accuracy and robustness, and adapting to laser chip defect detection under conditions of few samples.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HUNAN NORMAL UNIVERSITY
- Filing Date
- 2026-06-15
- Publication Date
- 2026-07-17
AI Technical Summary
Existing semi-supervised frameworks struggle to distinguish between highly reflective backgrounds and submicron-level defects in laser chip defect detection, leading to feature overload and noise accumulation, which affects detection accuracy and robustness.
We employ a multi-granularity attention residual module to separate features and perform adaptive aggregation. We also combine the topological dependencies between network layers to dynamically update parameters, suppress noise propagation, and improve detection accuracy and robustness.
It effectively separates highly reflective background features, preserves the local edges and fine structures of defects, reduces dependence on pixel-level annotation data, and improves the accuracy and stability of laser chip defect detection.
Smart Images
Figure CN122415602A_ABST