Anomaly detection during milk processing by autoencoder neural network based on near-infrared spectroscopy

Abstract

Abstract Anomaly detection during milk processing (such as changes in fat or temperature, added water or cleaning solution) can assure a satisfactory final product quality, including compositional and hygienic characteristics, as well as adulteration with water. The use of near-infrared (NIR) spectroscopy for change detection in complex dairy matrix is discussed. The autoencoder neural network plays fundamental role in anomaly detection. To evaluate this capability, the raw spectra obtained from NIR as well as first derivative and combination of both were analysed. An autoencoder was trained by 1.5% fat UHT-milk (measured at 5 °C) and applied to detect possible changes happening during the milk processing. The trained autoencoder using first derivative spectra was capable to detect 5% added water and 9% cleaning solution in the milk. Also, with the combination spectra, it was able to recognize a difference of 0.1% in fat concentration. In addition, both procedures were able to detect different production methods (specific procedure of suppliers such as homogenization level or pressure) and difference of 10 °C in the temperature. It can be concluded, that using an autoencoder neural network in combination with near-infrared spectroscopy is a reliable method to monitor the milk processing. By doing so, abnormal changes can be detected early, controlling the process becomes easier and the quality and safety of the product is guaranteed.