A color channel based on multiple Random Forest coupled with a computer vision technique for the detection and prediction of Sudan dye‐I adulteration in turmeric powder

Abstract

AbstractAdulterants can cause different health hazards upon prolonged consumption, but it is difficult to detect with human eyes. Non‐destructive turmeric adulteration detection is a challenging research area. The existing adulteration detection processes are largely instrumental and analytical with high accuracy but include limitations like long testing time, expensiveness, and lack of mobility. This work reports a new computer vision framework, which can simultaneously detect the presence of adulteration and predict the possible percentage of adulteration addressing the stated limitations. The scope has been remained to screening of Sudan dye‐I adulteration in turmeric powder. An in‐house database prepared with images of pure and adulterated turmeric powder samples has been used for experimentation. Random Forest algorithm has been employed for both classification and prediction. The model has been validated with standard internal and external validation methods to assess the stability and generalization potential of the model to avoid over‐ and under‐fitting problems. The results of classification show that the presented framework can provide more than 99% accuracy in detection while high correlation coefficient (R2) value in the tune of .99 for prediction. The novelty of the work is its simple histogram‐based color feature extraction, development of ensemble Random Forest prediction model that resulted in high accuracy and development of a faster, non‐invasive, less‐expensive, and validated screening method for adulterated turmeric powder that can be considered as a potential immediate screening method in the supply chain of powdered spices prior to confirmatory testing methods following two‐tiered food fraud testing approach.