Government-ID-Classifier

A deep learning project for automatic classification of government ID cards using transfer learning with VGG16. The model is fine-tuned on a custom dataset sourced from Roboflow and achieves an impressive 97% accuracy.

---

🚀 Features

• Uses VGG16 pretrained on ImageNet.
• Transfer Learning: all layers frozen except the last 4 for fine-tuning.
• Data Augmentation: applied for better generalization.
• High Accuracy: 97% on test data.
• Training, validation, and testing pipelines included (train.py, testing.py).
• Generates accuracy/loss plots and a confusion matrix for analysis.

---

📂 Repository Structure

Government-ID-Classifier/
├── train.py              # Training script (VGG16 + fine-tuning)
├── testing.py            # Testing script for inference and evaluation
├── requirements.txt      # Project dependencies
├── models/               # Directory to save trained weights
├── training_data/        # Training dataset 
├── validation_data/      # Validation dataset
└── testing_data/         # Testing dataset

🏋️ Training

Run the training script:

python train.py

• Uses VGG16 backbone with last 4 layers trainable.
• Trains for 35 epochs with data augmentation (rotation, shift, flip, etc.).
• Saves best weights as .keras files inside models/.
• Outputs plots:
  • model_accuracy.png
  • model_loss.png
  • confusion_matrix.png

---

🧪 Testing / Inference

Run the testing script:

python testing.py

• Loads the best saved weights (e.g., weights-improvement-02-0.98.keras).
• Evaluates accuracy on unseen test images.
• Prints classification report and lists misclassified images.

---

📊 Results

• Accuracy: ~97%
• Precision/Recall/F1: High across all classes
• Confusion Matrix: Saved as confusion_matrix.png
• Example evaluation snippet:

  ✅ Overall Test Accuracy: 97.00%
  

---

📜 License

This project is open-source. You may use, modify, and distribute it for research and educational purposes.

---

📌 Future Improvements:

• Add support for additional ID document types.
• Deploy as a web app using Flask/FastAPI.
• Experiment with newer architectures (e.g., EfficientNet, ViT).