Standard photographs convolutionally processed to indirectly detect gluten in chickpea flour

Abstract

In this work, an image-based method has been developed to detect the presence of traces of wheat flour in a gluten-free product such as chickpea flour, which may affect the health of the final consumers. For this purpose, a set of ground chickpea samples have been mixed with low amounts of wheat flour ranging from 1 to 50 ppm. Specifically, 12 groups of samples containing 1, 2, 3, 4, 5, 7.5, 10, 15, 20, 30, 40, and 50 ppm were prepared and photographed, along with the pure wheat and chickpea samples, with a regular camera to obtain 1400 images. Using a residual neural network, it was possible to accurately classify the samples into their specific group enabling a gluten detecting and indirectly quantifying tool. By performing a classification of a set of blinded samples, an overall accuracy greater than 93% was attained, validating the approach as a straightforward quality control and safety method. This prototype opens the door to simple and reliable health guarantees for consumers with celiac disease or gluten sensitivities.