Food powders classification using handheld Near-Infrared Spectroscopy and Support Vector Machine

Abstract

The aim of this study is to build a classifier model based on spectra data collected using handheld spectrometer that can classify between different types of food powders (flour and starch). A total of 70 samples were prepared from three different types of flour (whole wheat, organic wheat, and rice flour) and two different types of starch (corn and tapioca starch). Handpalm size handheld spectrometer is used to record the spectrum of each sample, the spectrometer has wavelength range of 900nm to 1700nm. The spectra data is pre-processed using gaussian smoothing to filter the data from noise and unrelated information. Multivariable data analysis method as principle component analysis (PCA) is used to eliminate irrelevant data and reduce the number of variables to three principle components for easier analysis and visualization. Support vector machine (SVM) is used to build a classification model. The training/calibration of the model was done by using 80% of the dataset while the remaining 20% was for testing the model. The results show that with proper pre-processing and PCA, classification of 100% accuracy can be achieved. This study indicates the potential future application of this approach for rapid detection in food powders fraud and adulteration.