Classification of adulterant degree in liquid solutions through interferograms with machine learning

In this research, the use of machine learning techniques to classify optical interferometric images based on their intrinsic characteristics is proposed and demonstrated. Using unsupervised machine learning algorithms, interferogram images, obtained and captured from a DACPI interferometer, are successfully classified based on their fringe pattern characteristics, for 6 different concentrations of isopropyl alcohol in commercial rum. From three sets of samples, confusion matrices and classification accuracy are obtained, reaching an accuracy of 90.78%. The results obtained represent an effective alternative to evaluate the characteristics of optical interferograms without the use of phase extraction techniques. Furthermore, the robustness of the results obtained for the unsupervised techniques are promising for analyses using supervised techniques to improve the classification accuracy of interferograms.