Robust multi-surface phase-shifting interferometry based on artificial neural networks

To mitigate the influence of phase-shift errors in wavelength-tuning phase-shifting interferometry, it is essential to enhance the precision of the phase-shifting steps and develop algorithms that are less sensitive to such errors. Leveraging the robust recognition capabilities of neural networks, we propose an Artificial Neural Network Phase-shifting Algorithm (ANNPA) for phase-shifting interferometry, detailing the network's design and training methods while optimizing calculation steps. By specially designing the training dataset, we fundamentally suppress high-order phase-shift errors. Simulations validate the algorithm's excellent insensitivity to both linear phase-shift errors and random noise in phase-shifting. Finally, we conducted a phase-shifting interferometry experiment using a Fizeau interferometer and employed ANNPA for calculations, with results clearly demonstrating its ability to perform multi-surface phase-shifting interferometry under non-ideal phase-shifting conditions.