High-accuracy identification of interferograms between two vortex beams via deep learning without adequate experimental data

Abstract

The on-axis interference intensity patterns between a vortex beam and its conjugated beam can be used to measure the fractional topological charge of vortex beams. However, it is still challenging to efficiently recognize these intensity diagrams. On one hand, the difference of the patterns for adjacent modes with interval 0.1 is too subtle to be identified precisely. On the other hand, the interferograms are susceptible to undesirable experimental conditions such as the misalignment of the beams, the unequal arms of the interferometer and the deviation of splitting ratio of the beam splitters in the interferometer. Here, we propose a deep learning method to recognize these intensity diagrams with up to accuracy. In particular, our method has reference values for deep learning model training when there is not adequate experimental data.