Robust interferogram processing using deep learning and signal subspace method for phase derivative estimation

Abstract

For non-destructive deformation metrology using optical interferometry, the derivative of phase map encoded in the interferogram signal contains crucial information about physical quantities such as displacement derivatives and strain. Hence, reliable retrieval of phase derivative is of great practical significance in precision metrology. However, this information is often difficult to retrieve in the presence of severe noise and imaging artifacts such as non-uniform intensity variations. In this paper, we propose a deep learning assisted signal subspace approach for extracting phase derivatives. The main advantages of the proposed method include robustness against severe noise and tolerance against interferogram abnormalities. The performance of the proposed method is validated using rigorous numerical simulations. The practical utility of the method is shown via experimental results obtained in digital holographic interferometry.