Comparative Analysis of Different Architectures of Deep Convolutional Neural Networks
Download| Volume 3 Issue 1 2022 | |
|---|---|
| Author(s): |
Mahwish Umer Mansoor Ahmed Khuhro Kamlesh Kumar Asif Ali Laghari |
| Abstract | Deep Neural Network (DNN) contains many layers of neurons, which provides better performance in terms of accuracy over a large dataset of images based on the DNN's artificial sense of recognition. The Deep Convolutional Neural Network (CNN) architecture is based on multiple convolutional layers of neurons and each neuron in a layer is designed towards feed-forward direction. CNN can learn features of unstructured data such as images, voice and videos during a model's training process. If the number of weighted layers increases in a CNN model, it results in higher object detection accuracy. On the other hand, if more weighted parameters are included in the CNN, it requires a high-performance GPU in the training session. The Xception model is better than VGG-16, ResNet50 and DenseNet121 because it passes the same input to the depth-wise isolated blocks and later merges the output of these blocks as input for the final classification layer. The model sizes are directly proportional to the number of parameters involved, affecting the models' performance during the object recognition process. These all models are pre-trained and require transfer learning for fine-tuning in a new dataset. There is a limitation to the models that the dataset must be in the RGB- system. This study compares the CNN architectures of VGG-16, ResNet50, Xception and DenseNet121. This paper's findings show that the Xception model of CNN has performed well while classifying digital images. |
| Keywords | Deep Convolutional Neural Network, VGG-16, ResNet50, Xception, DenseNet121 |
| Year | 2022 |
| Volume | 3 |
| Issue | 1 |
| Type | Research paper, manuscript, article |
| Recognized by | Higher Education Commission of Pakistan, HEC | Category | Y | Journal Name | ILMA Journal of Technology & Software Management | Publisher Name | ILMA University | Jel Classification | -- | DOI | - | ISSN no (E, Electronic) | 2709-2240 | ISSN no (P, Print) | Country | Pakistan | City | Karachi | Institution Type | University | Journal Type | Open Access | Manuscript Processing | Blind Peer Reviewed | Format | Paper Link | https://ijtsm.ilmauniversity.edu.pk/arc/Vol3/i1/pdf/4.pdf | Page |