Data Enhancement of Lens Defect Based on Dual Channel Generative Adversarial Networks

Fig. 4. Dataset acquisition system and results. (a) Machine vision lens defect detection system; (b) pretreatment of segmented lenses; (c) defect labeling

Download full size

Fig. 5. Results of lens defect image generated by DualC-GAN

Download full size

Fig. 6. Comparison of generation of lens defect images

Download full size

Table 1. Annotation of defect types
View table
Table 1. Annotation of defect types
No. Defect Label Count
1 Bubble Bubble 332
2 Scratch Scratch 303
3 Spot Spot 325
4 Smudge Smudge 318

Table 2. Quality evaluation of each model
View table
Table 2. Quality evaluation of each model
Method 1-NN MMD WD
CGAN 0.78 0.30 4.36
WGAN-GP 0.72 0.27 3.52
Patch GAN 0.61 0.18 3.12
DualC-GAN 0.52 0.15 2.81

Tools

Get Citation

Copy Citation Text

Qi Meng, Hua Miao, Lin Li, Bo Guo, Tingting Liu, Shilong Mi. Data Enhancement of Lens Defect Based on Dual Channel Generative Adversarial Networks[J]. Laser & Optoelectronics Progress, 2021, 58(20): 2015001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites