Indian Currency Recognition Web App

This is a Flask-based web application that uses deep learning models (MobileNetV2 and ResNet50) to recognize and classify Indian currency notes.

Features

• Upload an image of an Indian currency note.
• Real-time classification using pre-trained models.
• Support for multiple denominations (10, 20, 50, 100, 200, 500, 2000).

Setup & Installation

Important

Python Version Requirement: This project requires Python 3.9 to 3.12. (TensorFlow is currently not fully compatible with Python 3.13+).

1. Clone the repository:

   git clone https://github.com/albinmmathew/CurrencyWebApp.git
   cd CurrencyWebApp

2. Create a virtual environment:

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies:

   pip install -r requirements.txt

4. Run the application:

   python app.py

Training New Models

If you want to retrain the models or create new ones, follow these steps using the scripts in the root directory or the training/ folder.

1. Dataset Preparation

• Download the dataset from Kaggle: Indian Currency Note Images Dataset (2020)
• Extract the downloaded zip file into your root Project folder.
• IMPORTANT: Rename the extracted folder to simply dataset.
• Your Local Project Structure should look like this:

Project/ (Local Root Folder)
├── dataset/                     # Kaggle Dataset (Download & Rename)
│   └── Indian currency dataset v1/
├── class_indices.txt            # Class Mapping
├── train_mobilenet.py           # Training script (MobileNetV2)
├── train_resnet.py              # Training script (ResNet50)
├── compare_models.py            # Comparison & Confusion Matrices
├── requirements.txt             # Dependencies for training
└── CurrencyWebApp/              # Git Repository
    ├── training/                # Scripts (Copy to root folder to retrain)
    │   ├── train_mobilenet.py
    │   ├── train_resnet.py
    │   ├── compare_models.py
    │   └── requirements.txt     # Dependencies for training
    ├── models/                  # Latest .h5 model files
    ├── static/                  # UI Assets (CSS, JS)
    ├── templates/               # HTML Templates
    ├── app.py                   # Flask Application
    ├── requirements.txt         # Dependencies
    └── class_indices.txt        # Mapping file copy

2. Retraining & Evaluation

Retraining should be done in the Root Folder where the dataset/ is located.

Steps:

1. Copy Necessary Files: Copy the following files from CurrencyWebApp/training/ to your root folder:

   • train_mobilenet.py
   • train_resnet.py
   • compare_models.py
   • requirements.txt
   • class_indices.txt

2. Activate Virtual Environment: Ensure you are in the project folder and your venv is active:

   # On Windows:
   venv\Scripts\activate
   # On macOS/Linux:
   source venv/bin/activate

3. Install Dependencies:

   pip install -r requirements.txt

4. Run Training:

   python train_mobilenet.py  # For MobileNetV2
   # OR
   python train_resnet.py     # For ResNet50

5. Evaluate & Compare:

   python compare_models.py

   This will generate visual reports (.png files) and classification reports (.txt files) in the root folder.

3. Deploying New Models

• Copy the newly generated .h5 files into the models/ directory.
• Restart the Flask server to use the updated models.

Project Structure

• app.py: Main Flask application.
• models/: Contains pre-trained Keras model files.
• static/: CSS and client-side JavaScript.
• templates/: HTML templates.
• uploads/: Temporary storage for uploaded images.