基于DSDN-YOLO的PCB表面缺陷检测方法
By constructing a DSDN-YOLO model and utilizing DADC, SCGSFM, and DNI modules to enhance feature representation and fusion, combined with SA-Loss optimized regression, the problems of texture interference and uneven localization of small targets in PCB surface defect detection are solved, achieving high-precision defect detection.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SUQIAN COLLEGE
- Filing Date
- 2026-04-07
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies struggle to effectively suppress texture interference in complex backgrounds during PCB surface defect detection. This leads to the dilution of unstructured micro-defect discrimination features, blurred boundaries of slender defects, and uneven regression of small targets, all of which affect positioning accuracy.
We construct a deformable attention-based dynamic convolution module (DADC) to reconstruct C3k2 units, introduce a spatial-channel collaborative guided fusion module (SCGSFM) and a dynamic structure-aware interpolation strategy (DNI), and design a size-aware regression loss (SA-Loss) to enhance the network's ability to perceive small targets and geometric deformations, suppress background texture interference, and improve the expression of slender defect boundaries.
It improves the discriminativeness and stability of feature fusion in complex scenarios, enhances the positioning accuracy and detection accuracy of minute defects, and adapts to the inspection needs of high-density and miniaturized PCBs.
Smart Images
Figure CN122415463A_ABST