Edge-AI Implementation for Milk Adulteration Detection

Milk is an important source of nutrition consumed by most of the world's population. The introduction of adulterants (i.e. starch, sucrose, formaldehyde, etc) into milk, impacts quality as well as can cause severe health problems in the people consuming it. The use of the Internet of Things (IoT) with Artificial Intelligence (AI) can help predict and classify milk adulterants in real-time thereby informing dairy processors about the quality of milk during intake to the plant from dairy farms. The Edge-AI-based architecture allows the implementation of in-situ milk adulteration detection techniques in dairy processing, hence providing real-time monitoring of milk quality. The proposed architecture in this paper uses an edge device (Jetson Nano) to process the Fourier Transformed Infrared (FTIR) based data to classify different adulterants present in the milk dataset. A Convolutional Neural Network (CNN) model is used to address the classification problem. The edge device achieved 94.87% accuracy in classifying the adulterants present in the milk dataset. The model is successfully trained on the edge device providing benefits such as data governance, and flexible options for customizing the machine learning model at the edge in real time based on variations in adulterants.