Fig. 1. (a) Image collection schematic diagram; (b) Experimental device platform

Fig. 2. (a) High lighting schematic; (b) Low lighting schematic

Fig. 3. (a) Image of coating surface cracks; (b) Image of local cracks under microscope

Fig. 4. Comparison of low-recognition crack areas between 15°-60°. (a) 15°; (b) 30°; (c) 45°; (d) 60°

Fig. 5. Relationship between ray incidence angle and image contrast

Fig. 6. Image preprocessing flow chart

Fig. 7. Different image preprocessing methods. (a) Wavelet reconstruction; (b) Threshold segmentation; (c) Edge detection; (d) Proposed method

Fig. 8. (a) Original image and three-dimensional gray image; (b) Preprocessed image and three-bit gray image

Fig. 9. Data sets classification

Fig. 10. Flow chart of detection algorithm

Fig. 11. Improved U-net network structure diagram

Fig. 12. Schematic diagram of CBAM structure

Fig. 13. Schematic diagram of Depthwise over-parameterized convolution structure

Fig. 14. (a) Positive and negative mark masks; (b) Accuracy confusion matrix

Fig. 15. (a) Loss value change graph; (b) Accuracy change graph

Fig. 16. Training set segmentation results, each model is: (a) GT; (b) Op-U-net; (c) U-net; (d) Do-U-net; (e) Att-U-net; (f) Att-Do-U-net; (g) Reference[26]; (h) Reference[27]

Fig. 17. Independent test set segmentation results, each model is: (a) GT; (b) Op-U-net; (c) U-net; (d) Do-U-net; (e) Att-U-net; (f) Att-Do-U-net; (g) Reference[26]; (h) Reference[27]

Fig. 18. Comparison of segmentation results