Contrastive Augmented Transformer with Domain-specific Enhancement for Robust Multi-scenario Metal Surface Defect Detection
Researchers have developed a new framework called Contrastive Augmented Transformer (CAT) to improve the detection of metal surface defects in industrial manufacturing. This framework utilizes a hierarchical Swin Transformer backbone and an enhanced feature pyramid network to better identify subtle and multi-scale defects. To improve its performance in real-world conditions, CAT incorporates a domain-specific droplet augmentation algorithm and a hard negative mining strategy for contrastive loss. Experiments on the KolektorSDD2 dataset showed CAT achieved a 99.54% pixel-level AUROC, demonstrating strong generalization across various unseen datasets. AI
IMPACT This model could improve quality control in manufacturing by enabling more accurate and robust detection of subtle metal surface defects.