Indian Currency Recognition System

Project Overview

Production-ready Indian currency note recognition system using MobileNetV2 transfer learning. Identifies ₹10, ₹20, ₹50, ₹100, ₹200, ₹500 notes from images with 90-95% accuracy using just 200-600 images per class. Features a Streamlit web app with real-time predictions, confidence scores, probability charts, and downloadable reports.

Key Features

• 6 Indian denominations: ₹10, ₹20, ₹50, ₹100, ₹200, ₹500
• 90-95% accuracy with minimal training data (50 images/class recommended)
• 10x faster training (5-10 mins vs 15-20 mins for scratch CNNs)
• Real-time Streamlit app with image upload, predictions, and analytics
• MobileNetV2 optimized preprocessing (224x224, [-1,1] scaling)
• Production features: Confidence scoring, probability charts, JSON reports

Tech Stack

Category	Technologies
Deep Learning	TensorFlow 2.20+, Keras 3.11+, MobileNetV2
Web App	Streamlit
Computer Vision	OpenCV, PIL
Data Processing	NumPy, Pandas
Visualization	Matplotlib

Features:

• Upload currency image → Get instant prediction
• Confidence score + probability distribution chart
• Download detailed prediction report
• Mobile-friendly responsive design

Quick Start

1. Clone & Setup

git clone https://github.com/ImRAryan/IndianCurrencyPredictor.git
cd IndianCurrencyPredictor

2. Install Dependencies

pip install -r requirements.txt

3. Download Trained Model

# Model files (auto-downloaded by app)
best_mobilenetv2_currency_model.h5
label_mapping.json

4. Run Streamlit App

streamlit run app.py

requirements.txt

streamlit==1.38.0
tensorflow==2.20.0
keras==3.11.3
opencv-python==4.10.0
numpy==1.26.4
pandas==2.2.2
matplotlib==3.9.2
pillow==10.4.0
scikit-learn==1.5.1

Model Training (Optional)

1. Prepare dataset: Organize images in currencydataset/train/{10,20,50,100,200,500}/
2. Run training notebook: Indian_Currency_Recognition_MobileNetV2.ipynb
3. Minimum: 30 images/class (80-85% accuracy)
4. Recommended: 50 images/class (90-95% accuracy)