MeDERT: A metal surface defect detection model

Defects in various products are unavoidable because of measurement errors and equipment accuracy limitations the production process. Recent advances metal surface defect detection have focused on optimizing traditional methods, developing new techniques, exploring deep learning-based algorithms, providing technological support to improve manufacturing quality efficiency. To ensure highest yield rate meet requirements, all must undergo inspections before leaving factory. However, Traditional methods for detecting defects require a lot manual involvement, difficult accurately detect small defects, susceptible environmental interference, lack stability reliability. address this issue, we propose MeDERT model detection. Our approach involves Span-sensitive Texture Fusion (STF) module structure that focuses multi-headed attention modules recover lost details boost inspection speed top use Jump-sensitive detail recovery feature fusion validity extracted textures. Additionally, introduce singular value decomposition pretzel noise enhance robustness through data augmentation. achieved state-of-the-art (SOTA) results specified dataset, demonstrating its effectiveness enhancing efficiency accuracy.