Fig. 1. Schematics of round hole detection principle. (a) Front view; (b) left view

Fig. 2. Theoretical diffraction patterns of two circular holes with radius r under different δ (monochromatic light)

Fig. 3. Example of detection ranges for two wavelengths. (a) λ₁; (b) λ₂

Fig. 4. Schematic of MobileNetV3 network

Fig. 5. Hard Swish activation function and its derivative function

Fig. 6. Performance indexes of neural network. (a) Accuracy of validation set during network training; (b) prediction result of test set

Fig. 7. Relationship between number of test samples and residual RMSE and test accuracies under different R₂

Fig. 8. Relationship between number of test samples and residual RMSE and test accuracies under different signal-to-noise ratios

Fig. 9. Residual RMSEs and test accuracies under different R₂ and S_SNR

Fig. 10. Detection results for piston errors under different detection methods