Measurement of mirror piston misalignments for large segmented telescopes. Diffraction images are generated to train neural networks. The goal is to quantify non-smooth wavefront phase shifts. Various deep learning approaches are used: CNN, RNN and reinforcement learning.
