Improving CIFAR-10 Image Classification with a Modified ResNet-18 and Advanced Data Augmentation Strategies
DOI:
https://doi.org/10.66849/JIADS.v1i0.6Keywords:
Convolutional Neural Networks, CIFAR-10;, Image Classification;, Data Augmentation;Abstract
Image classification tasks on low‑resolution datasets such as CIFAR‑10 involve significant challenges due to the limited spatial information available and the heightened risk of model overfitting when employing deep convolutional neural networks that are originally engineered for high‑resolution images. This paper presents an optimized deep learning
system based on an adapted ResNet-18, that is, specifically designed to classify images efficiently on the CIFAR-10 dataset. To retain spatial resolution in 32x32 images, the standard 7x7 convolutional stem of ResNet -18 is replaced by a 3x3 convolutional layer and the opening max -pooling operation is not used. Also, in order to boost generalization, more powerful approach to data-augmentation that includes random cropping, horizontal
flipping, AutoAugment policies and random erasing is carried out during the training. The AdamW optimiser, which includes label smoothing, is used to perform model optimisation, and cosine annealing learning-rate schedule used to ensure that the learning rate smoothly converges. The experimental findings indicate that the suggested method has a test
accuracy of 92.75 00 per cent on the CIFAR-10 dataset. The steady increase in the validation performance during all training epochs suggests high generalization ability. Therefore, the suggested framework provides a practical and efficient approach to classification performance improvement on low-resolution image datasets.
References
[1] Y. Liang, “Enhancing educational insights through convolutional neural networks: A comparative analysis on CIFAR-10 and CIFAR-100 datasets,” in Proc. Int. Conf. Intelligent Education and Artificial Intelligence, 2024.
[2] A. Krizhevsky and G. Hinton, “Learning multiple layers of features from tiny images,” University of Toronto, Toronto, ON, Canada, Tech. Rep., 2009.
[3] F. Giuste and J. Vizcarra, “CIFAR-10 image classification using feature ensembles,” arXiv preprint arXiv:2002.03846, 2020.
[4] K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in Proc. IEEE Conf. Computer Vision and Pattern Recognition (CVPR), 2016, pp. 770–778.
[5] S. Shin and J. Kim, “Implementation and comparison of CNN models on CIFAR-10 dataset,” in Proc. Int. Conf. Industrial Engineering and Operations Management, 2024.
[6] M. Tan and Q. Le, “EfficientNet: Rethinking model scaling for convolutional neural networks,” in Proc. Int. Conf. Machine Learning (ICML), 2019.
[7] N. Srivastava, G. Hinton, A. Krizhevsky, I. Sutskever, and R. Salakhutdinov, “Dropout: A simple way to prevent neural networks from overfitting,” Journal of Machine Learning Research, vol. 15, pp. 1929–1958, 2014.
[8] P. Thanapol, S. Hongsongkiat, and P. Jirawimut, “Reducing overfitting and improving generalization in training CNN under limited sample sizes in image recognition,” in Proc. IEEE Int. Conf. Computer Science and Engineering, 2020.
[9] W. Shi, Y. Gong, and J. Wang, “Improving convolutional neural network performance with min-max objective,” in Proc. Int. Joint Conf. Artificial Intelligence (IJCAI), 2016.
[10] V. Vasilev, D. Dimitrov, and P. Petrov, “Optimizing activation function for parameter reduction in convolutional neural networks on CIFAR-10 dataset,” Applied Sciences, vol. 15, no. 8, pp. 1–15, 2025.
[11] S. Singla, S. Singla, and S. Feizi, “Improved deterministic robustness with 1-Lipschitz neural networks,” arXiv preprint arXiv:2002.03846, 2020.
[12] A. M. Ionescu et al., “A mixed-signal binary convolutional neural network processor for energy-efficient inference,” in Proc. IEEE Int. Solid-State Circuits Conf. (ISSCC), 2023.
[13] E. D. Cubuk, B. Zoph, D. Mane, V. Vasudevan, and Q. V. Le, “AutoAugment: Learning augmentation policies from data,” in Proc. IEEE Conf. Computer Vision and Pattern Recognition (CVPR), 2019.
[14] Z. Zhong, L. Zheng, G. Kang, S. Li, and Y. Yang, “Random erasing data augmentation,” in Proc. AAAI Conf. Artificial Intelligence, 2020.
[15] Y. Abouelnaga, O. S. Ouda, and M. H. B. Hassan, “CIFAR-10: KNN-based ensemble of classifiers,” in Proc. Int. Conf. Computational Science and Computational Intelligence, 2016.
