Automatic Identification of Teff (Eragrostis Tef) Flour Adulteration Using Machine Learning

Abstract

Food adulteration, the act of mixing foodstuffs with lower-quality foodstuffs or other food and non-food items, has become a global issue, particularly in developing countries like Ethiopia. This is a significant problem as it can lead to nutritional deficiencies and in some cases, illnesses, diseases or even death. The identification of adulterated food typically requires physio-chemical laboratory tests that vary depending on the type and level of adulterant. Teff flour, a daily staple food in Ethiopia, is often adulterated with wood flour and gypsum powder. In this research work, we utilized computer vision techniques to extract textural and color features from images of both adulterated and pure Teff flour. We then trained a Support Vector Machine (SVM) based machine learning model to identify adulterated Teff flour and predict the level of adulteration. The presented results show varying levels of prediction accuracy depending on the type of Teff flour, the level of adulteration. The developed model was also found to provide good results with images sizes ranging from <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\boldsymbol{128 \times 128}$</tex> to <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\boldsymbol{2048 \times 2048}$</tex> pixels. The results show overall model accuracy was above 90% when the adulteration level of red Teff flour is more than 25% and above 84% when the adulteration level of white wood flour was above 25%.