Specialized CNN Architectures for Enhanced Image Classification Performance
S. M. Deepa, J. Jayapriya, M Vinay, P. Kalpana, J Roselin Suganthi, J. Loveline Zeema · 2024
Image classification is one of the important tasks in computer vision, with a greater number of applications from facial recognition, medical imaging, object recognition and many more. Convolutional Neural Networks (CNNs) have developed as the foundation for image all classification tasks, showcasing the capacity to learn the hierarchical features automatically. In this study proposed three custom CNN models and its comprehensive analysis for the image classification tasks. The models are evaluated using CIFAR-10 dataset to assess the performance and efficiency. The experimental results shows that the proposed custom CNN Model-3 performance is better than the other two models. Our findings demonstrate that Model 3, featuring with the global average pooling, achieves the highest overall accuracy of $94 \%$ with competitive computational efficiency. This suggests that global average pooling is the valuable technique for balanced and accurate image classification.