Classification of Food Powders with Open Set using Portable VIS-NIR Spectrometer

Near Infrared (NIR) spectroscopy is fast and non-destructive methods for analyzing materials without pretreatment. Especially as portable NIR spectrometers have been developed, the research of spectral analysis has applied to various open environment and field. In this paper, we classify visually indistinguishable eight food powders using portable VIS-NIR spectrometer with a wavelength range of 450 to 1000 nm with CNN (Convolutional Neural Network), one of the machine learnings. Further we consider open set recognition where unknown classes should be rejected at test time. The proposed CNN model achieved an accuracy of 100% for eight food powders, and 91.2% with open set. Our experimental results demonstrate the potential of material analysis using a portable VIS-NIR spectrometer with machine learning.