Fast demodulation of single-shot interferogram via convolutional neural network.

In this report, a method is proposed to demodulate a single-shot interferogram based on deep learning using a convolutional neural network (CNN). The time required to demodulate a single interferogram is on the order of milliseconds. The root mean square (RMS) of the residual wavefront between the original wavefront and the estimated wavefront reached the order of 10−2 λ. Moreover, the cost was lower than using commercial dynamic and phase-shifting interferometers. This method yielded superior real-time performance and high precision. The results indicated that this approach based on deep learning is robust and feasible for demodulating single-shot interferograms. • A method based on deep learning is proposed to demodulate a single shot interferogram. • The single shot interferogram is used as the input of the convolutional neural network, and the Zernike polynomials coefficient is output. • Compared with other methods of demodulating a single shot interferogram, the method is faster, and the speed of demodulating an image is in the order of microseconds. The RMS (Root-Mean-Square) of the residual wavefront between the original wavefront and the estimated wavefront approached 10 − 2 λ. • The cost of the method is lower than commercial dynamic and phase-shifting interferometers. In addition, we compare the speed and accuracy of the method with the Zygo interferometer, and find that the method leads to better or comparable results yet is orders of magnitudes faster.it will be great helpful for the dynamic fringe demodulation or other real-time applications. • The process of demodulation of the convolutional neural network is robust. [ABSTRACT FROM AUTHOR]