Table 5 Comparative experimental results of models on the NEU-DET dataset.
From: DEENet: an edge-enhanced CNN–Transformer dual-encoder model for steel surface defect detection
Methods | mAP/% | Precision/% | Recall/% | F1-score/% | Param/M | FLOPs/G |
|---|---|---|---|---|---|---|
Faster RCNN | 60. 6 | 77.9 | 76.3 | 78.9 | 60. 1 | 246.4 |
SSD | 72. 4 | 79.3 | 84.8 | 82.0 | 25. 0 | 64.2 |
YOLOv5s | 70. 3 | 78.5 | 78.5 | 81.2 | 7. 2 | 27.7 |
YOLOv9 | 73. 7 | 79.6 | 79.9 | 80.8 | 12. 1 | 32.9 |
YOLOv10 | 71. 8 | 80.3 | 81.4 | 81.3 | 8. 0 | 40.6 |
YOLOv11 | 73. 8 | 79.9 | 79.8 | 80.0 | 9. 4 | 42.8 |
RT-DETR40 | 75.0 | 79.3 | 81.4 | 79.9 | 42 | 136 |
MSD-YOLO41 | 80.9 | 83.2 | 82.4 | 84.9 | 35.3 | 54.2 |
MD-YOLO42 | 78.2 | 82.6 | 81.6 | 82.1 | 9.0 | 14.1 |
DEENet | 81.4 | 84.8 | 85.6 | 85.2 | 8.2 | 12.4 |
