Fast and high precision phase recovery technology of single-shot ineterferogram based on depth convolution neural network

Abstract

Phase retrieval from single interferogram is of great interest for the possibility of dynamic phase measurement. However, it is a very complicated process in reality since the unknowns and knows are highly unequal. In this paper, we propose a fast phase recovery method from single interferogram based on deep convolution neural network. The network is trained based on supervised learning to achieve the purpose of quickly obtaining unwrapped phase results from a single interferogram. To improve the detection accuracy, a modified set establishment model has been propose to improve the practicability of the fringe data. The simulation and experimental results show that the root mean square value of residual phase extraction error by this method is closed to 0.01λ (λ = 632.8 nm), and the constructed depth convolution neural network model has significant flexibility and effective generalization ability for phase recovery of single frame interference fringe.