Fast phase distortion identification and automatic distortion compensated reconstruction for digital holographic microscopy using deep learning

Abstract

Digital holographic microscopy (DHM) is a quantitative phase measurement technique with full-field, contactless, and fast. The technique provides accurate micro-surface morphology of samples. These steps are essential for accurate phase reconstruction, such as holographic focusing, numerical diffraction, phase unwrapping and distortion compensation. Performing these processes manually is time-consuming and is not conducive to the general application of the technology. In order to improve the detection efficiency, this paper proposes a deep learning model that can achieve fast identification of DHM phase distortion and automatic phase distortion compensation reconstruction. The model can be preprocessed for holographic phase to accurately identify the type of phase distortion present in the phase. And adaptively adjust the network weight parameters for phase distortion compensation reconstruction. The experimental results show that the method proposed in this paper achieves fast and accurate identification of multiple phase distortions. The model has high accuracy and strong generalization ability. The reconstructed holographic phase map has PSNR of 35.2743dB and RMSE as low as 10^-2 level in the face of complex mixed aberrations. The identification and reconstruction processes took 0.005s and 0.058s, both in milliseconds, respectively. The evaluation indexes SSIM, FSIM and NC can reach above 0.99. It is shown that the method in this paper is not only capable of reconstructing holograms, but also able to effectively retain the detailed features of the original image.