Artificial Intelligence Aided Adulteration Detection and Quantification for Red Chilli Powder

Food adulteration imposes a significant health concern on the community. Being one of the key ingredients used for spicing up food dishes. Red chilli powder is almost used in every household in the world. It is also common to find chilli powder adulterated with brick powder in global markets. We are amongst the first research attempts to train a machine learning-based algorithms to detect the adulteration in red chilli powder. We introduce our dataset, which contains high quality images of red chilli powder adulterated with red brick powder at different proportions. It contains 12 classes consists of 0%, 5%, 10%, 15%, 20%, 25%, 30%, 35%, 40%, 45%, 50%, and 100% adulterant. We applied various image color space filters (RGB, HSV, Lab, and YCbCr). Also, extracted features using mean and histogram feature extraction techniques. We report the comparison of various classification and regression models to classify the adulteration class and to predict the percentage of adulteration in an image, respectively. We found that for classification, the Cat Boost classifier with HSV color space histogram features and for regression, the Extra Tree regressor with Lab color space histogram features have shown the best performance.