End-to-End Direct Phase Retrieval from a Single-Frame Interferogram Based on Deep Learning

TY - JOUR

T1 - End-to-End Direct Phase Retrieval from a Single-Frame Interferogram Based on Deep Learning

AU - Zhang, Tianshan

AU - Lu, Mingfeng

AU - Hu, Yao

AU - Hao, Qun

AU - Wu, Jinmin

AU - Zhang, Nan

AU - Yang, Shuai

AU - He, Wenjie

AU - Zhang, Feng

AU - Tao, Ran

N1 - Publisher Copyright: © 2024 IEEE.

PY - 2024

Y1 - 2024

N2 - In the field of optical interferometry, phase retrieval is a critical step in acquiring the phase information of fringes. Various deep-learning-based methods have been proposed for phase retrieval from a single-frame interferogram. However, the existing methods still cannot obtain the unwrapped phase directly without the aid of extra steps. To truly fulfill end-to-end phase retrieval for various fringe patterns, we propose a novel method with carefully crafted network architecture and training methodology. Experimental results on simulated and actual interferograms show excellent accuracy, noise robustness, and demodulation efficiency without any further phase unwrapping or polynomial fitting required by the existing methods. Furthermore, the proposed method is compatible with non-Zernike-polynomial-phase interferograms containing phase discontinuities. These properties have qualified the proposed method for high-standard interferometric measurement for optical fabrication.

AB - In the field of optical interferometry, phase retrieval is a critical step in acquiring the phase information of fringes. Various deep-learning-based methods have been proposed for phase retrieval from a single-frame interferogram. However, the existing methods still cannot obtain the unwrapped phase directly without the aid of extra steps. To truly fulfill end-to-end phase retrieval for various fringe patterns, we propose a novel method with carefully crafted network architecture and training methodology. Experimental results on simulated and actual interferograms show excellent accuracy, noise robustness, and demodulation efficiency without any further phase unwrapping or polynomial fitting required by the existing methods. Furthermore, the proposed method is compatible with non-Zernike-polynomial-phase interferograms containing phase discontinuities. These properties have qualified the proposed method for high-standard interferometric measurement for optical fabrication.

KW - Deep learning

KW - fringe analysis

KW - optical interferometry

KW - optical metrology

KW - phase retrieval

UR - https://www.scopus.com/pages/publications/85197545944

U2 - 10.1109/TIM.2024.3418112

DO - 10.1109/TIM.2024.3418112

M3 - Article

AN - SCOPUS:85197545944

SN - 0018-9456

VL - 73

JO - IEEE Transactions on Instrumentation and Measurement

JF - IEEE Transactions on Instrumentation and Measurement

M1 - 7004316

ER -