Abstract:

In the present study, we put forward an algorithm based on support vector machine (SVM) for fast identification
of different types of flour by attenuated total reflectance-Fourier transform infrared spectroscopy (ATR-FTIR). ATR-FTIR
spectra of 139 samples of four common types including strong flour, wheat core flour, refined snowflake flour and bread
flour, were collected randomly. The outlier samples were eliminated based on Mahalanobis distance and an SVM model was
established to predict samples. The binary tree SVM model was used to identify the types of flour, and the parameters of the
kernel function were optimized by using the grid method .The results showed that the recognition accuracy reached 100%,
100%, 75% and 85.71% for strong flour, refined snowflake flour, wheat core flour and bread flour, respectively, and the
average recognition accuracy of the model was 90.177 5%. All the above results indicate that it is feasible to use ATR-FTIR
with SVM algorithm for quick and accurate identification of different types of flour.

Key words: flour, ATR-FTIR, Mahalanobis distance, support vector machine