Research Article
LE-YOLOv5: A Lightweight and Efficient Road Damage Detection Algorithm Based on Improved YOLOv5
Table 5
Stepwise ablation experiment.
| Models | Backbone | Modules | Algorithms | Params (M) | mAP0.5 (%) | ImpV3 | GSConv | VCACSP | PAFF | K-means | Label smoothing |
| Baseline | | | | | | | 7.20 | 51.6 | Case 1 | ✓ | | | | | | 4.37 | 52.0 | Case 2 | ✓ | ✓ | | | | | 4.03 | 52.3 | Case 3 | ✓ | | ✓ | | | | 3.72 | 52.1 | Case 4 | ✓ | | | ✓ | | | 4.37 | 53.3 | Case 5 | ✓ | ✓ | ✓ | | | | 3.38 | 53.6 | Case 6 | ✓ | ✓ | | ✓ | | | 4.03 | 54.6 | Case 7 | ✓ | | ✓ | ✓ | | | 3.72 | 54.1 | Case 8 | ✓ | ✓ | ✓ | ✓ | | | 3.41 | 56.0 | LE-YOLOv5 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | 3.41 | 56.9 |
|
|
✓ means the modules or algorithms are used.
|