Deep learning based method for phase analysis from a single closed fringe pattern

Interferometry is a widely used optical measurement technique. We can estimate the physical parameters of the measured object by analyzing the phase of the fringe pattern obtained by interference imaging. However, when the measurement object has spherical surface, the interferogram always contains closed fringes which the traditional analysis methods are difficult to handle. Therefore, we use several common deep learning networks to learn the closed fringe patterns and their phases, evaluate and choose the appropriate network to build an end-to-end phase analysis system for a single closed fringe pattern. The experimental results show that the constructed deep learning network model has excellent phase recovery effect on simulation closed fringe patterns, and can estimate the curvature radius of the spherical surface accurately.