Milk adulteration poses a global concern, with developing countries facing higher risks due to unsatisfactory monitoring systems and policies. Surprisingly, this common issue has often been overlooked in many countries. Contrary to popular belief, adulterants in milk can result in severe health risks, potentially leading to fatal diseases. Detecting and categorizing milk adulteration is crucial for consumer safety and the dairy industry. This research is divided into 2 breakthroughs, destructive and non-destructive methods. In the destructive method, the Lactoscan system was used for qualitative analysis: (Solid Not Fat (SNF), density, fat, lactose, conductivity, solids, protein, temperature, and pH level). The research also examines non-distractive hyperspectral imaging (HSI) through HSI Specim Fx-10 (397-1003 nm) analysis to detect various phases of milk adulteration for accurate and user-friendly imaging-based adulterants detection and categorization. Preprocessing involves radiometric correction, image resizing, region of interest (ROI) selection for feature extraction, and empirical line method (ELM) to calculate spectral reflectance signature. Machine learning techniques (Logistic Regression (LRs), Decision Tree (DTs), Support Vector Machine (SVMs), and Linear Discriminant Analysis (LDAs)), are employed, with LDA excelling in adulteration identification by learning the spectral signatures. These algorithms are trained and validated using a developed milk adulteration data set. Training, testing, and validation accuracy, precision, recall, F1-score, Kappa, and Matthew's correlation coefficient (MCC) metrics showcase the effectiveness of the proposed pipeline, outclassing numerous state-of-the-art approaches with a validation accuracy of 100%. In conclusion, this study established a multiclass model capable of detecting milk adulterant behavior, showing significant practical application for milk quality assessment.
Read full abstract