A hybrid quantum-classical approach for fruit classification and calorie prediction using machine learning

Abstract

In the evolving field of modern agriculture, the integration of machines and deep learning techniques, particularly for fruit classification, has become increasingly significant due to the diverse characteristics of fruits. Our research makes a substantial contribution by introducing a high-quality dataset of fruit images and conducting numerical experiments to train neural networks for fruit detection. This study explores the motivations behind focusing on fruit classification and the practical applications of the developed classifiers. We evaluated the effectiveness of various models, including convolutional neural networks (CNN), ResNet50, VGG19, and DenseNet201, by measuring their performance in terms of loss and accuracy on both test and validation datasets. The CNN model outperformed the others, achieving a test loss of 0.1580 and an accuracy of 96.63%, alongside a validation loss of 0.1020 and an accuracy of 97.15%. ResNet50, VGG19, and DenseNet201 also demonstrated promising results, though slightly less accurate than CNN. These findings underscore the efficacy of these models in precisely categorizing fruits and quantifying their caloric content, signifying substantial progress in agricultural technology and food science. Additionally, the hybrid quantum-classical model yielded encouraging outcomes for both objectives, achieving an overall accuracy of 98.2%. © 2026 Walter de Gruyter GmbH. All rights reserved.