Fig. 1
From: On the utilization of deep and ensemble learning to detect milk adulteration
a Plot of the infrared spectra for three randomly selected samples of the classes pure, formaldehyde and peroxide, acquired by the FTIR equipment. The raw spectra were analyzed directly by the proposed Convolutional Neural Network. Each spectrum was plotted with subtle shifts for viewing purpose. b Component features for the same samples, generated by the FTIR equipment and stored in CSV format. Each column quantifies an important milk composition information. The columns for fat, protein, lactose, solids, solids non-fat (SNF), casein and milk urea nitrogen (MUN) represent each component concentration in the sample. The Cells column represents the somatic cells counting, FrzPoint represents the freezing point of the sample, with values given in degrees Hortvet (∘H), and QValue is a calculation of the sample quality by the equipment. These numerical features were analyzed by ensemble methods