Achieving High Accuracy in Lemon Quality Classification: A Comparative Study of Deep Learning and Transformer Models

Abstract

Abstract Agricultural product quality assessment is important for the efficiency and marketability of production. Quality assessment improves industry standards, increasing sales and reducing crop loss. Maintaining quality is of paramount importance for all processes, from production to sales. Artificial intelligence has recently been frequently used for product quality assessment in the agricultural field. Both in the literature and in practice, deep learning and machine learning methods are used to process images of agricultural products and evaluate their quality. They are classified according to specified standards. In this study, firstly, data augmentation operations were performed on the lemon dataset consisting of two classes, bad quality and good quality, by using rescaling, random zoom, random flip, and random rotation methods. Afterward, eight different deep-learning methods and two different transformer methods were used for classification. As a result of the study calculated the most successful result as 99.84% accuracy, 99.95% recall, and 99.66% precision with the ViT method. This value is the highest accuracy value in the literature. When the experimental results are evaluated, it shows that lemon classification processes are successfully performed using the ViT method.