基于改进yolo11的白坯布缺陷检测研究

Key Points

• The modified DAM-YOLO model achieved a 2.3% increase in mAP@0.5 compared to the original YOLO11 model, indicating improved defect detection accuracy.
• Enhanced feature extraction utilized the C3K2-DCNv2 module, which increased adaptability to deformations and complex backgrounds.
• The integration of adaptive scale fusion and DySample modules allowed dynamic sampling and spatial adaptation, improving precision for small target detection.
• The MultiSEAMHead reinforced target representation through multi-branch semantic decoupling and context aggregation strategies.

Abstract