Measurement of visibility and phase steps of a static wind imaging interferometer assisted by deep learning.

ABSTRACT

In a static wind imaging Michelson interferometer we developed, one of the Michelson mirrors is divided into four quadrants, with coatings on the quadrants that provide small phase steps from one quadrant to another, realizing the four simultaneous sampling of the interferogram. Restricted by the coating process and interferometer adjustment, the instrument visibility and phase steps of the four quadrants will deviate from the design value. In the actual passive detection of the atmospheric wind field, quasi-real-time calibration is required, and the calibration will also be affected by the instrument noise. In this paper, we propose a deep-learning-based denoising algorithm that can quickly denoise the wind interferogram with no need to adjust parameters, combined with conventional least-squares fitting cosine curves to obtain the visibility and phase steps of four quadrants from a series of interferograms with varying phase differences. The proposed algorithm framework is verified by experiment, and the measurement of visibility and phase steps of the wind field interferogram is efficiently realized. It can provide a reference for the visibility and phase steps measurement of the wind imaging interferometer and may have applications in wind imaging interferometer calibration.