Fig. 1. Gray images of metal collected from multiple angles of light. (a) From the right; (b) from the bottom; (c) from the left; (d) from the top

Fig. 2. Schematic of photometric stereo hardware. (a) Hardware layout diagram; (b) light source vector calibration diagram

Fig. 3. Schematic for solving normal vector of highlighted points

Fig. 4. Surface normal map reconstructed by photometric stereo method

Fig. 5. Model network structure with super division ratio of 2

Fig. 6. In the process of upsampling on the network, the elements in the four feature layers correspond to the new feature layer graph

Fig. 7. Flow chart of proposed method

Fig. 8. Quantitative super-resolution reconstruction test platform. (a) Experiment platform; (b) shooting result

Fig. 9. Imaging detailed display of gray image. (e) Blade gray image; (b)-(e) corresponding local area pictures

Fig. 10. Gray scale images collected from different light source angles

Fig. 11. Detailed comparison of quantitative super-resolution reconstruction. (a) Blade normal map; (b)-(e) corresponding local area pictures

Fig. 12. Comparison of feature boundaries before and after fusion reconstruction. (a) Fig. 9 (b) area; (b) Fig. 11 (b) area

Fig. 13. Comparison of boundary extraction effects

Fig. 14. Comparison of high-resolution and low-resolution images in real images. (a) Original image; (b) local high-resolution image; (c) local down sampling low-resolution image

Fig. 15. Comparison of the effect of different super dividing networks on the surface image of metal blades. (a) Original picture; (b) local picture; (c)-(g) effects of local area using RDN,SRGAN,EDSR,SRCNN,and proposed method

Fig. 16. Metal blade image and surface defects. (a) Original picture; (b) local picture

Fig. 17. Strengthening results of different methods. (a) Method of Ref. [10]; (b) method of Ref. [5]; (c) proposed method

Fig. 18. Defect detection results by different methods. (a) Original detection image; (b) detection result by method of Ref. [10];

Fig. 19. Effect of micro defect boundary extraction after image super-resolution enhancement and reconstruction. (a) Local picture boundary extraction; (b) enhanced reconstruction feature extraction; (c)(d) detail enlarged view