Deep-learning approach to measuring the refractive index of transparent liquids

Abstract

A deep-learning approach is introduced to determine the refractive index of transparent liquids based on variations in the displacement of ultra-smooth interference fringes. The phase characteristics of these fringe variations captured in video data were analyzed and modeled using group-phase fitting. A neural network model, integrating a dense convolutional network with a long short-term memory network, was then developed and trained for high-precision liquid refractive index measurements. Experiments demonstrated an R2 accuracy of 99.70% and a mean squared error of 0.0003. This methodology has been confirmed to be temperature-dependent, considerably stable against external disturbances, highly accurate, and capable of real-time processing.