Autoencoder-based clustering of DACPI interferograms for detection of adulteration levels of two different dopants in alcoholic beverages

In this research, a dense autoencoder for clustering (AEC) is proposed as a hybrid learning method to classify interferograms obtained from a dual-aperture common-path interferometer due to their intrinsic patterns, which represent the degree of adulteration of ketamine and isopropyl alcohol (IPA) as dopants in commercial rum. The AEC is capable of identifying and classifying single-shot interferograms for the two different dopants at a low computational cost and high accuracy, using two different error metrics to evaluate the reconstruction quality: the logarithmic mean square error and the Theil-U coefficient. The highest classification accuracy was obtained for the Theil-U metric, reaching 0.9947 for IPA concentrations and 0.9968 for ketamine. The AEC is demonstrated as a reliable artificial intelligence method to support interferometric-based optical instrumentation applications.