Single-frame dual-wavelength phase demodulation method based on deep learning

In dual-wavelength interferometry, how to accurately separate single-wavelength interferogram from a dual-wavelength interferogram and retrieve the high-precision phase distribution from a single frame interferogram is a critical problem. In order to solve this problem, a single frame dual-wavelength interferogram modulation method based on deep learning strategy is proposed in this paper. Dual-wavelength network (D-Net) and Phase network (P-Net) are proposed in this study. The method only requires a single frame dual-wavelength interferogram. With a well-trained D-Net network, the interferograms corresponding to the two different wavelengths can be extracted from the single frame dual-wavelength interferogram respectively, and are taken as the input of P-Net. P-Net outputs the wrapping phase of the two wavelengths, and then the dual wavelength phase can be obtained by unwrapping the wrap phase. Finally, the tested optical element surface shape can be obtained from the phase distribution. Furthermore, instead of using real experimental data, an interferogram generation model is constructed to generate the dataset for the network's training. The learning rate attenuation strategy adopting appropriate optimizers and loss functions is introduced to guarantee the high-accuracy training of the network. Simulations have been done to validate the feasibility of this algorithm. The simulations prove that this method can guarantee a high detection accuracy and expand detection range, which provides a solution for the phase recovery problem in dual wavelength interferometry.