A deep neural network and random forests driven computer vision framework for identification and prediction of metanil yellow adulteration in turmeric powder

Turmeric (Curcuma longa) is a popular food ingredient which is widely used in powdered form. Despite different food and medicinal advantages it is often adulterated. Metanil yellow (MET) is one such synthetic chemical which can be easily mixed with turmeric powder and such mixing is difficult to detect. This paper presents a computer vision framework using the potential of deep neural network towards detection of MET adulteration in turmeric powder and random forests regressor to predict the possible amount of adulterant. An in-house database consisting of features from turmeric images of five variants of pure and adulterated turmeric powder has been used for experimentations. A new frequency domain annular-mean filter-based feature extraction has been used. The results show the potential of the presented method that can perform with more than 98% accuracy in both identification and prediction tasks. The reported technique can be considered as a motivating step towards development of a non-invasive and low-cost mobile device towards food adulteration detection in future.