Fig. 1. YOLOv4 network structure

Fig. 2. Improved YOLOv4 network structure

Fig. 3. Type of defect. (a) Crack; (b) gap; (c) pit; (d) scratch

Fig. 4. Data enhancement diagram. (a) Original picture; (b) horizontal flip; (c) exposure adjustment; (d) Mosaic data enhancement

Fig. 5. Data labeling diagram. (a) Original picture; (b) picture annotation example; (c) xml tag file

Fig. 6. Loss function curve

Fig. 7. Various defect detection results. (a) (e) Crack; (b) (f) gap; (c) (g) pit; (d) (h) scratch

Fig. 8. Comparison of the AP of different algorithms under the original YOLOv4 network

Fig. 9. Comparison of the F_β of different algorithms under the original YOLOv4 network

Fig. 10. Comparison of the AP of different networks under improved parameter adjustment algorithm

Fig. 11. Comparison of the F_β of different networks under improved parameter adjustment algorithm