A Wavelet, Fourier and PCA Data Analysis Pipeline:  Application to Distinguishing Mixtures of Liquids

Using a new optical engineering technique for the “fingerprinting” of beverages and other liquids, we study and evaluate a range of features. The features are based on resolution scale, invariant frequency information, entropy, and energy. They allow mixtures of beverages to be very precisely placed in principal component plots used for the data analysis. To show this we make use of data sets resulting from optical/near-infrared and ultrasound sensors. Our liquid “fingerprinting” is a relatively open analysis framework in order to cater for different practical applications, in particular, on one hand, discrimination and best fit between fingerprints, and, on the other hand, more exploratory and open-ended data mining.