Determination of margarine adulteration in butter by machine learning on melting video

Butter is a product that is often vulnerable to adulteration with cheaper ingredients such as margarine. In this study, butter was artificially adulterated with margarine at different rates to create different levels of adulteration. Then, the melting was captured using video footage, and image processing and machine learning (ML) were used to automatically detect the level of adulteration in the butter. To create the final numerical dataset for ML models, a total of 30,000 images were collected from the video, with equal numbers of images for each class. The images were divided into five classes using an algorithm that detected region of interest (ROI) in the adulterated butter images. Two types of numerical datasets were created: single frame-based and first-middle-last (FML) frame-based. Seven different ML models (decision tree (DT), linear discriminant analysis (LDA), Naïve Bayes (NB), support vector machines (SVM), k-nearest neighbor (KNN), random forest (RF) and artificial neural networks (ANN) were trained and tested on the datasets. To improve accuracy and efficiency, 10-fold cross-validation was applied to the ML models. The ML models achieved high accuracy in classifying the loaded butter videos. KNN, RF, and ANN had the highest accuracy (99.9%), followed by SVM (99.7%) and DT (99.4%) on the single frame-based dataset. NB had the lowest accuracy (87.1%). On the FML frame-based dataset, DT had the highest accuracy (99.9%) while SVM had the lowest accuracy (73.3%). Overall, the method used in this study was successful in classifying butter adulteration with high accuracy using image processing and ML techniques.

Graphical Abstract