Milk Quality Inspection Using Hyperspectral Imaging

Milk has been an essential part of our food culture as it contains important micronutrients and macronutrients. Milk is contaminated by the addition of water and preservatives. Traditionally, screening of milk quality was performed using human-based methods which have limitations such as being labor-intensive, time-consuming, and expensive. Therefore, non-destructive testing of milk quality using Hyperspectral imaging (HSI) is implemented. Compared to manual milk quality tests, HSI (Hyperspectral image) is faster and does not involve destructive methods. Pasteurized milk and vendor milk are used for sample preparation whereas water, Ammonium sulphate, and Ammonium chloride are chosen as adulterants. Therefore, the database is generated by capturing the images of milk samples with three different types of adulterants that are mixed with milk (Water, Ammonium Sulphate, and Ammonium Chloride) using the Resonon Hyperspectral camera (pika L, 400–1000 nm). Further, they are classified into three class classifications depending on the level of adulterants added. The problem of feature redundancy and noise is solved by using PCA-based Explained variance. On choosing ROI, the mean spectral curve is obtained and the optimal wavelength is chosen for extracting features and trained through machine learning classifiers like Ensemble, K-nearest neighbor, and Support Vector Machine for the three-class classification problem out of which the K-nearest neighbor, classifier reported the highest accuracy of 87%, 85%, 88% for vendor milk adulterant level classification and 84%, 87%, 85% for pasteurized milk adulterant level classification.