Edible Oil Adulteration Detection Using Deep Learning

Edible oil adulteration is an important food-quality concern because it can affect product authenticity and consumer confidence.This paper presents an image-based approach for edible oil classification and adulteration assessment using deep learning techniques.Four deep learning models, namely MobileNetV2, DenseNet121, EfficientNetB0, and Vision Transformer (ViT), were evaluated using a custom image dataset containing sunflower, soybean, mustard, and peanut oil samples.The reported training accuracies were 95.40%, 97.74%, 94.97%, and 97.22%, respectively.The best reported validation accuracies were 100%, 100%, 99.31%, and 98.61%, respectively.Although DenseNet121 and MobileNetV2 achieved higher validation accuracy in the reported experiments, ViT was selected for integration into the project application.A webbased system was developed to allow users to upload oil images and view the model's prediction.The proposed system demonstrates how computer vision can support preliminary edible oil screening.Further testing with independent samples and laboratory-based verification is required to establish its reliability for practical use.