Chinese Journal of Liquid Crystals and Displays, Volume. 40, Issue 9, 1356(2025)
Urban road defect detection algorithm based on improved YOLOv8n
Fig. 8. Detection results before and after algorithm improvement. （a） Transverse crack； （b） Longitudinal crack； （c） Repairment.
Fig. 9. Grad-CAM visualization results before and after algorithm improvement. （a） Small-scale defects； （b） Damages under complex background； （c） Multi-scale defects.
Fig. 10. Comparison of experimental results. （a） Multiple transverse cracks； （b） Repairment； （c） Multiple types of cracks.
Table 1. Experimental results of different attention mechanisms
View tableView in ArticleTable 1. Experimental results of different attention mechanisms
YOLOv8n Parameters/M GFLOPs Precision/% ReCall/% mAP50/% mAP（50～95）/% +ECA 3.006 8.1 91.2 90.0 93.8 68.7 +SimAM 3.006 8.1 93.3 89.4 93.8 68.6 +LSKA 3.077 8.1 95.3 89.1 94.5 69.0 +CAA 3.207 8.4 94.1 89.9 94.7 70.0 +MLPA（Ours） 3.221 8.3 95.9 91.2 96.4 71.4
Table 2. Experimental results of different loss functions
View tableView in ArticleTable 2. Experimental results of different loss functions
YOLOv8n Parameters/M GFLOPs Precision/% ReCall/% mAP50/% mAP（50～95）/% +Focal_DIoU 3.006 8.1 92.8 89.7 94.6 69.4 +EIoU 3.006 8.1 93.1 90.1 93.5 68.7 +MPDIoU 3.006 8.1 93.2 90.2 94.1 68.6 +PIoU 3.006 8.1 94.2 89.8 94.6 70.5 +WIoU（Ours） 3.006 8.1 94.6 89.9 95.3 70.5
Table 3. Ablation study results
View tableView in ArticleTable 3. Ablation study results
Method MLPA C2f_D WIoU Precision/% ReCall/% Parameters/M GFLOPs mAP50/% mAP（50～95）/% YOLOv8n 93.2 91.3 3.006 8.1 94.3 70.5 A √ 95.9 91.2 3.221 8.3 96.4 71.4 B √ 95.0 92.0 3.203 8.5 96.8 74.9 C √ 94.6 89.9 3.006 8.1 95.3 70.5 D √ √ 96.0 94.3 3.417 8.7 98.2 77.8 E √ √ 94.6 93.3 3.221 8.3 96.9 72.6 F √ √ 94.5 91.0 3.203 8.5 97.6 75.7 G √ √ √ 96.8 96.0 3.417 8.7 98.5 78.1
Table 4. Comparison of experimental results for different algorithms
View tableView in ArticleTable 4. Comparison of experimental results for different algorithms
Methods Parameters/M GFLOPs mAP50/% mAP（50～95）/% YOLOv7n 37.21 105.1 96.2 83.1 YOLOv7-tiny 6.20 13.2 94.0 69.2 YOLOv7x 70.83 188.9 96.8 81.4 YOLOv9-t 60.80 266.1 95.5 73.7 YOLOv10n 2.70 8.2 93.0 67.8 YOLOv10s 8.04 24.5 94.4 71.9 YOLOv11n 2.58 6.3 94.2 68.3 FastestDet 0.24 1.7 71.4 41.6 Road-YOLOv5 3.85 8.3 95.4 76.8 Road-EfficientDet 6.58 6.0 93.0 53.6 YOLOv8-road 3.42 8.7 98.5 78.1
Get Citation
Copy Citation Text
Shisong ZHU, Hong GAO, Bibo LU, Haijing DU. Urban road defect detection algorithm based on improved YOLOv8n[J]. Chinese Journal of Liquid Crystals and Displays, 2025, 40(9): 1356
Category:
Received: May. 15, 2025
Accepted: --
Published Online: Sep. 25, 2025
The Author Email: Bibo LU (lubibo@hpu.edu.cn)
