Fig. 1. Structure of YOLOv8s network

Fig. 2. Structure of GSConv network

Fig. 3. Structure of C2f network and Bottleneck

Fig. 4. Structure of C2f-Faster network

Fig. 5. PConv working principle and FasterNet structure

Fig. 6. Structure of CF-EMA network

Fig. 7. Feature fusion structure after adding small defect detection layer. (a) Default path of YOLOv8s; (b) after adjusting

Fig. 8. Structure of GCP-YOLOv8s network

Fig. 9. Aerial images of partial insulators. (a) Display of defects; (b) examples of data enhancement; (c) fogging effect under different fog thickness (L: brightness, D: fog thickness)

Fig. 10. Information visualization of insulator defect dataset. (a) Category and quantity; (b) distribution of central points; (c) defect size distribution

Fig. 11. Comparison of mAP@0.5 of GCP-YOLOv8s and YOLOv8s

Fig. 12. Comparison of mAP@0.5 of each model

Fig. 13. Detection effect of YOLOv8s and GCP-YOLOv8s in different backgrounds