Vol. 12, Issue 11 (2023)

Author(s):
Deepoo Meena, Neeraj Tiwari, Shrankhla Mishra and Ravi Prakash Pandey

Abstract:
Assessing the quality of food and spices is crucial for ensuring human nutrition. Researchers have explored non-destructive methods, particularly computer vision, for measuring food and spice quality. This study emphasizes the significance of quality assessment for turmeric, given its high nutritional value and susceptibility to fraudulent practices. Low-quality wheat powder, due to its low market price, becomes an attractive option for adulterating turmeric powder. The research employs an improved convolutional neural network (CNN) to classify turmeric powder images and detect fraud. A dataset of 3000 image samples is categorized into six groups, encompassing pure turmeric powder and various levels of adulteration with wheat flour (10%, 15%, 20%, 25%, and 34%). The study aims to enhance fraud detection capabilities, contributing to the preservation of turmeric's integrity in the market. During the initial image processing step, undesired components were eliminated. Employing data augmentation (DA) was crucial to mitigate overfitting issues in the convolutional neural network (CNN). Specifically, the VGG-16 (Visual Geometry Group 16 layer) model architecture was deployed for the classification task. In implementing this deep learning approach, the image dataset was randomly partitioned into two primary sets: 90% for the training-validation phase of the CNN and 10% designated for a blind test. The refined model exhibited an impressive 92.7% accuracy during the validation phase, with a minimal 5.6% misclassification rate observed in blind testing. This underscores the efficacy of the method as a quality and safety control measure for the turmeric industry. The study outcomes also underscore the potential of computer vision, particularly in conjunction with deep learning (DL), as a valuable tool for assessing quality and uncovering fraudulent practices in turmeric powder.