Improved VGG Architecture in CNNs for Image Classification

Apart from computer vision, deep learning has brought the concept to a new era of machine learning. One of the deep learning approaches for classification analysis is Convolutional Neural Networks (CNNs), a model of artificial neural network that has often been the most popular approach in computer vision. In recent decades, many approaches for image classification have been proposed. To obtain high accuracy, most studies focused on deepening and enlarging the CNNs architecture such as the VGG network. However, deep and complex architecture, on the other hand, can result in extraordinarily long execution time. This study primarily aims to classify images using the improved VGG architecture to minimize the execution time and enhance the classification performance. The comparative experiments of the proposed architecture with another three existing architectures have been made and trained with six different datasets from Kaggle. As a result, the execution time and the classification accuracy of the proposed architecture is better than the other three existing architecture. Hence, the proposed architecture indicates that the execution time and the classification performance can be improved by downsized the VGG architecture.