Detecting white interference fringes in noisy conditions via supervised learning neural networks

Abstract

White interferograms are signals recorded using the interference of light with positional information obtained using a white interferometer, which is a laser technology that uses light interference for the non-contact measurement of the surface topography of a sample. In long-range measurements, the reflected light weakens with increasing distance. When the peak-to-peak value of interference fringe signals is less than the noise level interference fringes cannot be visually confirmed. We propose a neural network-based supervised learning method to detect white interference fringes in noisy conditions. The interference fringes obtained were transformed into time series signals and stored as images. Some of the data was used to train the neural network. Some data was used to validate the trained neural network. The trained model could distinguish between the presence and absence of white interference fringes in noise-contaminated conditions with a certain probability. Numerical calculations and optical experiments validated the proposed method. This technique can be applied to detect weak reflections and extend the interferometry measurement range.