Deep Transfer Learning for Improved Quality Control in Organic Tomatoes

The current study explores the effective employment of Transfer Learning (TL) based on pretrained ALexNet Convolutional Neural Network (CNN) model for classifying organic tomato images at different stages throughout the food supply chain, aiming to enhance quality control in agrifood supply chain domain. Images of healthy, infected, ripe, and unripe fresh organic tomatoes were captured at real environment conditions and categorized as fresh healthy-diseased, fresh ripe-unripe and packaged ripe-unripe corresponding to three distinct checkpoints throughout the supply chain, namely harvesting, sorting, and packaging. High accuracies after TL employment, have been achieved including 93,44 % for detecting fresh healthy versus fresh diseased tomatoes, 82,54 % for fresh ripe versus fresh unripe tomatoes, and 97,5% for ripeness in packaged tomatoes. The attained results indicate the potential of the presented deep TL method to enhance food supply chain management through the accurate assessment of quality attributes in organic agrifood products.