Deep-learning-based rapid wavefront sensor using a line-scan camera

We present a rapid wavefront sensor based on machine learning and using a line-scan camera. The object light wave propagates through a scattering medium. In our method, the scattered light wave undergoes a series of preconditioning steps. The resultant light wave, in which only the wavefront aberration component is emphasized and the reference object light wave is removed, is captured as one-dimensional data using line focusing optics. The captured data are trained by a convolutional neural network, and the trained network can estimate the Zernike coefficients without iterative calculations. The proposed method achieves significantly faster measurement compared to a two-dimensional sensor. The proposed method was experimentally demonstrated, as a proof of concept, using a line-scan camera and a preconditioning method that we designed.