One-frame interferometric surface contouring via stepwise phase extraction and deep learning

• Proposing a one-frame deep learning method for accurate phase extraction. • Eliminating sign ambiguity, harmonics, and phase-shift errors simultaneously. • Utilizing synthetic noisy data to improve robustness and generalization. • Demonstrating superior accuracy over conventional multi-frame PSA methods. • Enabling compact, fast, and scalable interferometric measurements. Precise inspection of interferometric optical flats is essential as the performance of the final applications directly depends on the surface characteristics. We propose a novel deep learning-based one-frame phase extraction framework for precise surface contouring of optical flats in optical interferometry. By integrating the strengths of spatial and temporal methods, the framework effectively addresses harmonic distortion, phase-shift errors, and sign ambiguity using only one experimental interferogram. Interferogram generation and phase extraction are executed sequentially using simulation datasets and trained models. Experimental validation on an optical flat demonstrates superior accuracy and robustness compared with conventional multi-frame methods. This approach significantly reduces experimental limitations while enhancing scalability in intelligent high-precision interferometric measurements.