Research on Image Recognition Based on Different Depths of VGGNet

Authors:
Open Access
Publisher website
Google Scholar
Abstract
With the advancement of image recognition technology, the significance of convolutional neural networks has continually increased. The VGGNet model, developed by the Visual Geometry Group at the University of Oxford, has proven successful, demonstrating that the depth of the network is crucial for model performance. This study aims to explore the impact of various depths of VGGNet models on image recognition tasks. Three classic VGG network models were selected: VGG-13, VGG-16, and VGG-19, along with two widely used image datasets, MNIST and CIFAR-10, for a comprehensive experimental analysis. The experimental results on the CIFAR-10 dataset indicated that as network depth increased, there was a significant enhancement in model accuracy, with VGG-19 performing the best. This outcome confirms the superiority of deep networks in processing complex image data. Conversely, in the simpler MNIST dataset, the deeper VGG-19 did not exhibit better performance compared to VGG-13, suggesting that excessively deep networks might not be necessary for simple datasets and could lead to overfitting and gradient vanishing.