Fig. 1. Overall network architecture for the improved YOLOv5 algorithm

Fig. 2. Schematic of 3×3 DCNv2 layer

Fig. 3. Input and output feature maps for DCNv2 module

Fig. 4. Diagram of CA Module

Fig. 5. Input and output feature maps for CA module

Fig. 6. Structure of CA-PANet

Fig. 7. Fusion feature map of CA-PANet

Fig. 8. Raw image as well as vertically flipped， horizontally flipped， contrast enhanced and brightness enhanced images

Fig. 9. Raw EL image

Fig. 10. Comparison of AP and mAP between YOLOv5 before improvement and our YOLOv5

Fig. 11. Defect image detection with the addition of different improvement modules

Fig. 12. Defect detection results for different models of the same dataset

Fig. 13. Improved YOLOv5 model for common defect detection results