Fused dispersive convolution and spatial attention-based magnetic tile defect detection method and device
By introducing divergent convolution and spatial attention mechanisms into the magnetic tile defect detection model, optimizing the backbone network and feature fusion, the problems of low contrast and multi-scale defect detection are solved, and high-precision, lightweight magnetic tile defect detection is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HUNAN OPEN UNIV (HUNAN PROVINCIAL CADRE EDUCATION & TRAINING ONLINE COLLEGE)
- Filing Date
- 2026-04-26
- Publication Date
- 2026-07-17
AI Technical Summary
Existing magnetic tile defect detection models are insufficient in low-contrast and multi-scale defect feature extraction, have low fusion efficiency, and are difficult to achieve high-precision real-time detection on embedded devices, resulting in robustness and deployment adaptability issues.
A magnetic tile defect detection method that integrates divergent convolution and spatial attention is adopted. By replacing the standard convolutional layers of the backbone network with divergent convolutional layers, and combining the DFEB and SAM modules for feature fusion, the backbone network is optimized to enhance the feature extraction capability, and the computational load is reduced through a lightweight network architecture.
It improves the ability to capture features of low-contrast and multi-scale defects, enhances the model's adaptability and detection accuracy for small, medium and large-scale defects, reduces computational complexity, and is suitable for deployment in embedded devices.
Smart Images
Figure CN122415549A_ABST