Open Access

ARTICLE

Improving Date Fruit Classification Using CycleGAN-Generated Dataset

by Dina M. Ibrahim^1,2,*, Nada M. Elshennawy²

1 Department of Information Technology, College of Computer, Qassim University, Buraydah, 51452, Saudi Arabia
2 Computers and Control Engineering Department, Faculty of Engineering, Tanta University, Tanta, 37133, Egypt

* Corresponding Authors: Dina M. Ibrahim. Emails: ,

(This article belongs to the Special Issue: New Trends in Statistical Computing and Data Science)

Computer Modeling in Engineering & Sciences 2022, 131(1), 331-348. https://doi.org/10.32604/cmes.2022.016419

Received 04 March 2021; Accepted 11 October 2021; Issue published 24 January 2022

Abstract

Dates are an important part of human nutrition. Dates are high in essential nutrients and provide a number of health benefits. Date fruits are also known to protect against a number of diseases, including cancer and heart disease. Date fruits have several sizes, colors, tastes, and values. There are a lot of challenges facing the date producers. One of the most significant challenges is the classification and sorting of dates. But there is no public dataset for date fruits, which is a major limitation in order to improve the performance of convolutional neural networks (CNN) models and avoid the overfitting problem. In this paper, an augmented date fruits dataset was developed using Deep Convolutional Generative Adversarial Networks (DCGAN) and CycleGAN approach to augment our collected date fruit datasets. This augmentation is required to address the issue of a restricted number of images in our datasets, as well as to establish a balanced dataset. There are three types of dates in our proposed dataset: Sukkari, Ajwa, and Suggai. After dataset augmentation, we train our created dataset using ResNet152V2 and CNN models to assess the classification process for our three categories in the dataset. To train these two models, we start with the original dataset. Thereafter, the models were trained using the DCGAN-generated dataset, followed by the CycleGAN-generated dataset. The resulting results demonstrated that when using the ResNet152V2 model, the CycleGAN-generated dataset had the highest classification performance with 96.8% accuracy, followed by the CNN model with 94.3% accuracy.

---

Keywords

---