A PCB defect detection method based on robust statistics and dynamic weight of Bayesian network
By employing robust statistics and dynamic weighting of Bayesian networks, this method addresses the issues of sensitivity to outliers and fixed feature weights in traditional PCB defect detection. It achieves robust identification and efficient detection of six types of defects, thereby improving the robustness and accuracy of the detection system.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIJING NEW ENERGY VEHICLE TECH INNOVATION CENT CO LTD
- Filing Date
- 2026-04-15
- Publication Date
- 2026-07-17
AI Technical Summary
Traditional PCB defect detection methods are sensitive to outliers, and fixed feature weights lead to an imbalance in the identification of the six types of defects, resulting in a high rate of false positives and false negatives. They also lack a unified process for robust feature standardization and dynamic feature weight allocation.
We employ a method based on robust statistics and dynamic weights in Bayesian networks. We construct a sample-feature weighted covariance matrix by using HL/Shamos robust standardization and Tukey double quadratic kernel to calculate sample weights, perform dimensionality reduction of feature vectors, and perform defect posterior probability inference in a Bayesian network.
It improves the robustness and recognition accuracy of the detection system, reduces the cost of multi-model maintenance and manual re-judgment, and enhances the detection efficiency and process traceability of the production line.
Smart Images
Figure CN122415477A_ABST