Authors:
Fatma Çakıroğlu
1
;
Ali Durmuş
2
;
Ercan Karaköse
3
;
Rifat Kurban
4
and
Selami Parmaksizoğlu
5
Affiliations:
1
Kayseri University, Vocational School of Information Technologies, Kayseri, Turkey
;
2
Kayseri University, Faculty of Engineering, Arch. and Design, Dept. of Electrical & Electronics Eng., Kayseri, Turkey
;
3
Kayseri University, Faculty of Engineering, Arch. and Design, Dept. of Basic Sciences, Kayseri, Turkey
;
4
Abdullah Gül University, School of Engineering, Department of Computer Engineering, Kayseri, Turkey
;
5
Antalya Bilim University, School of Civil Aviation, Department of Pilotage, Antalya, Turkey
Keyword(s):
Data Augmentation, Deep Learning, Image Data, Big Data, CIFAR-10 Dataset.
Abstract:
Deep learning studies require large amounts of training data to prevent overfitting. However, labelled data is limited for real-world applications. Data augmentation can be performed by increasing the diversity and quantity of training data. As an effective way to increase the adequacy and diversity of training data, data augmentation has become an indispensable part of successfully applying deep learning models to image data. In this paper, a CNN model tailored to the CIFAR-10 dataset was employed. Basic data augmentation was performed using classical data augmentation methods such as random horizontal flipping, random rotation, and random scaling, and the models were retrained with the augmented dataset. The results obtained using the CIFAR-10 dataset were compared with those obtained using the augmented dataset. The results indicate that the work conducted with the augmented dataset yielded better performance and higher accuracy. This highlights the importance of data augmentation
in image processing with deep learning.
(More)