Mir-IR femtosecond laser pulses wavefront reconstruction using machine learning methods

The paper shows the possibility of using convolutional neural networks for a laser beam wavefront reconstruction by measuring the intensity distribution in the transverse plane at some distance behind the focusing system. The optimal distance from the focusing system to the measurement plane is determined at which the error for the rms devation of Zernike polynomials amplitudes decreases sharply. A neural network has been trained on a computer-generated samples of beams with various types of aberrations to reconstruct the wavefront of real beams in the presence of noise