Table 6 Comparative experimental results of models for each defect type on the NEU-DET dataset.
From: DEENet: an edge-enhanced CNN–Transformer dual-encoder model for steel surface defect detection
Methods | Cr | In | Pa | Ps | Rs | Sc |
|---|---|---|---|---|---|---|
Faster RCNN | 37.9 | 77.8 | 91.5 | 80.4 | 60.2 | 89.6 |
SSD | 38.7 | 76.8 | 88.5 | 78.0 | 65.4 | 77.4 |
YOLOv5s | 46.0 | 82.0 | 91.0 | 84.0 | 71.4 | 89.8 |
YOLOv9 | 46.2 | 80.1 | 95.4 | 80.0 | 72.2 | 91.2 |
YOLOv10 | 49.2 | 81.6 | 93.4 | 72.1 | 68.3 | 85.3 |
YOLOv11 | 44.4 | 81.7 | 94.8 | 82.1 | 70.5 | 93.6 |
RT-DETR40 | 45.5 | 85.7 | 91.8 | 83.7 | 67.8 | 91.3 |
MSD-YOLO41 | 56.3 | 84.3 | 92.0 | 83.1 | 72.3 | 97.7 |
MD-YOLO42 | 46.7 | 81.4 | 91.3 | 85.1 | 72.6 | 92.0 |
DEENet | 56.5 | 88.3 | 96.8 | 87.3 | 68.8 | 93.7 |
