Improving Visual Recognition with Enhanced Convolutional Neural Networks
Yamini Madhav Patil, Gunita Arun Chandhok, B. R. Supreeth, Sumit Pundir, Subramanyam Kunisetti, S. Gayathri · 2024
The study focuses on evaluating the performance of enhanced Convolutional Neural Network (CNN) architecture on the CIFAR-100 dataset for image recognition tasks. The proposed CNN architecture and all four models, namely ResNet50, DenseNet121, EfflcientNet-B3, and one, were trained and evaluated. The performance of the proposed CNN architecture is undeniably superior to that of the other models evaluated, with an accuracy of 99% and precision of 98% In terms of performance, DenseNet121, EfficientNet-B3, and ResNet50 all achieved commendable results, with an accuracy of 90% and precision of 91%. The effectiveness of the proposed convolutional neural network architecture in categorizing images in multiple dimensions demonstrates the potential for discrete visual recognition tasks. The results underscore the importance of custom CNN designs in real-world applications that require high precision and accuracy. The results of the paper offer valuable insights for future efforts to enhance visual recognition systems and push forward the field of computer vision.