Deep Learning for Automatic Classification of Carotenoid Associated Color Pigmentation

Abstract

This study explores the application of deep learning models, specifically ResNet-34, ResN et-50, and EfficientNet-B0, for the automatic classification of carotenoid-associated color pigmentation in tomatoes. The dataset comprises 250 images categorized into five pigmentation levels, reflecting the varying carotenoid content. Carotenoids, such as lycopene and beta-carotene, are key pigments influencing the color of tomatoes, with deeper reds and oranges indicating higher concentrations. The models were evaluated for direct classification and regression followed by classification. Results show that EfficientNet-B0 achieved the highest accuracy in direct classification (94.00%), while ResNet-34 excelled in regression tasks (91.33%). Future research will continue exploring regression tasks to predict actual carotenoid content in tomatoes, enhancing prediction accuracy and robustness.