Towards Spectral Data Reduction in Paprika Powder Adulteration Detection

This paper investigates a well-known food safety problem, namely the adulteration of paprika powder with brick powder. Recently, rapid and non-destructive methods for detecting adulteration using spectrography and hyperspectral imaging have been proposed, the major drawback of which is the large amount of data to be processed and stored. We investigate five simple feature selection methods for the reduction of spectral measurement data to simplify the measurement procedure and speed up data processing. Specifically, we apply the feature selection methods to spectral data of paprika powder and brick powder to identify the most informative spectral bands. The results show that the most informative spectral bands are mostly found in the ranges from 510 to 680 nm and from 760 to 920 nm, and they vary depending on the method used. All other spectral data could be omitted without significantly reducing and in some cases even increasing the classification capability of paprika and brick powder. The highest classification capability of 98.1% is achieved with a data reduction of 82 % and thus even exceeds the classification capability of the original, unreduced data set of 96.8%. The highest data reduction of 96.9% can be achieved by selecting only two spectral bands, resulting in a classification capability of 78 %. These results show the potential for simpler and more economical solutions for the detection of spice adulteration.