This project explores the use of Convolutional Neural Networks (CNN) for image recognition using the CIFAR-10 dataset. By leveraging transfer learning through fine-tuning the VGG-16 model, the student developed a high-accuracy classifier. The model achieved 85% accuracy and was optimized by adjusting the last three fully connected layers. A graphical user interface (GUI) built using the Tkinter library enables non-programmers to use the model for classification. Users can load a trained model, upload a 24-bit image, and receive classification results directly on the interface. This system aims to simplify the use of deep learning for general users while maintaining reliable recognition performance.
The modified VGG-16 model effectively classifies images in the CIFAR-10 dataset with high accuracy and reduced overfitting. Its integration into a GUI allows broader accessibility and usability for non-technical users. The application is suitable for educational and practical use in image classification tasks.
The model currently supports only 24-bit depth RGB images and may underperform on datasets with drastically different image characteristics. The reliance on VGG-16 and CIFAR-10 limits its adaptability without further training or domain-specific tuning.
Planned enhancements include expanding the model’s compatibility to different image formats and bit depths, integrating other pre-trained models for comparative analysis, and deploying the tool as a web application to enhance accessibility. Additional work may include support for edge computing environments.