A Custom CNN Architecture for Image Recognition using CIFAR‑10

Hayden Bin Nor Azman
Abdul Salam Shah

Journal of ICT, Design, Engineering and Technological Science
Volume 8, Issue 2

Deep Learning algorithms have remained prominent in image classification across various domains. CNNs have modernized the process, enabling automated classification in new fields. Traditional models suffer from lower accuracy largely due to manual feature extraction. To address challenges in modern algorithms, this paper proposes a CNN‑based model for the multi‑class problem using the CIFAR‑10 dataset. The hyperparameters have been carefully selected to achieve higher accuracy while avoiding overfitting. Data augmentation and dropout layers contributed to achieving 85.49% accuracy