This project implements a deep learning image classification system using a fine-tuned VGG19 model on the CIFAR-10 dataset. The system was trained to identify 10 classes of images, achieving a training accuracy of 99% and validation accuracy of 88%. The project further deploys a Flask-based GUI for real-time image recognition, showcasing practical application of AI in vision tasks. This confirms the strength of deep learning models, particularly convolutional neural networks, in real-world image recognition scenarios.
The study demonstrates that deep convolutional networks like VGG19 are effective in image classification tasks when properly trained and fine-tuned. The project highlights the real-world feasibility of applying pre-trained models in new domains through transfer learning and custom GUI deployment. It offers promising directions for machine vision applications in various sectors.
The model, while accurate, remains constrained by dataset noise and low resolution of CIFAR-10 images. Moreover, deployment is limited by the computational cost of real-time CNN inference and dataset-specific generalizability.
Future enhancements will explore the use of deeper ensemble architectures and higher-resolution datasets. Incorporating autoML tools and expanding the interface for multi-model comparison will be considered to improve scalability and accuracy.