Harsh0809/fruit-quality-classification-cnn

🍎 SavorShield - Fruit Freshness Classification System

SavorShield is an interactive deep learning web application that uses a Keras Convolutional Neural Network (CNN) to instantly classify fruits as Fresh or Rotten. It features a modern Flask-based Python backend and a premium glassmorphic dark-mode web user interface styled with Vanilla CSS and high-fidelity Javascript micro-interactions.

🚀 Live Demo on Hugging Face Spaces:

👉 [SavorShield Live Classifier](https://huggingface.co/spaces/Harsh0809/fruit-quality-classification-cnn)

🌟 Key Features

• —Instant CNN Inference: Pre-loads the 134MB Keras model (fruits_fresh_rotten_model.keras or fruits_fresh_rotten_model.h5) at server startup to deliver quick prediction times.
• —Compatibility Patching: Features a custom runtime patch to resolve Keras 2/3 serialization mismatches (quantization_config arguments) dynamically at startup, guaranteeing stable loading.
• —Drag-and-Drop Dropzone: Implemented with responsive drag events, instant client-side preview, and verification checks.
• —Dynamic Visual Themes: The interface adapts to predictions:
• —Vibrant Emerald theme for Fresh fruits.
• —Amber/Crimson theme for Rotten fruits.
• —Conic-Gradient Progress Gauge: Smoothly animates the confidence level returned by the sigmoid layer.
• —Latency & Architecture Stats: Displays neural network characteristics (Inference delay, model sequence layers, total latency, and exact sigmoid probabilities).
• —Single-Click Portability: Windows batch script wrapper (run.bat) to automate installation and host launching.

📂 Project Architecture

d:\Data science course\11-Deep learning\6-food_classification\
├── fruits_fresh_rotten_model.keras    # Pre-trained Keras Model file
├── fruits_fresh_rotten_model.h5       # Fallback Keras Model file
├── app.py                             # Flask Web Server & Inference Pipeline (with Keras 2/3 compatibility patches)
├── requirements.txt                   # Environment Dependencies (optimized for tensorflow-cpu)
├── run.bat                            # Windows Environment Setup & Launcher
├── Dockerfile                         # Deployment configuration for Hugging Face Spaces
├── README.md                          # Project Documentation
├── templates/
│   └── index.html                     # Semantic Web Structure
└── static/
    ├── css/
    │   └── style.css                  # Custom CSS Stylesheet & Glassmorphic variables
    └── js/
        └── main.js                    # Javascript Controller & API Fetches

⚙️ How It Works (The Inference Pipeline)

1. 1.Upload: The user uploads an image (PNG, JPG, JPEG, or WEBP) up to 16MB.
2. 2.Preprocessing: The Flask application opens the image stream via Pillow, converts it to RGB color channels, and resizes it to $224 \times 224$ pixels matching the model input shape requirement.
3. 3.Normalization: The pixel matrix values are scaled to `[0.0, 1.0]` by dividing by 255.0 to match model scaling during training.
4. 4.Execution: The input tensor is fed into the Keras model. The final layer uses a single-node dense sigmoid activation that outputs a probability value ($P$) between 0 and 1.
5. 5.Class Mapping:
6. 6.If $P < 0.5$: The fruit is classified as Fresh (Confidence = $(1 - P) \times 100\%$).
7. 7.If $P \ge 0.5$: The fruit is classified as Rotten (Confidence = $P \times 100\%$).
8. 8.Result Transformation: The JSON output triggers theme state overrides (.is-fresh / .is-rotten) in JavaScript, transitioning the UI context theme.

☁️ Hugging Face Spaces Docker Deployment

SavorShield is fully Dockerized and deployed under Hugging Face Spaces utilizing:

• —`python:3.10-slim` base image to keep image overhead minimal.
• —Standard Port `7860` for proper routing inside Space containers.
• —`tensorflow-cpu>=2.16.1` to leverage Keras 3 while remaining lightweight and avoiding out-of-memory errors.
• —Git LFS (Large File Storage) tracking rules configured to handle Keras model weight files larger than 10MB.

🚀 Installation & Local Launch

Option A: Double-Click Launcher (Recommended for Windows)

1. 1.Double-click the [run.bat](file:///d:/Data%20science%20course/11-Deep%20learning/6-food_classification/run.bat) file.
2. 2.It will automatically check for Python, verify dependencies (installing Flask, Pillow, and Numpy if missing), and start the server.

Option B: Terminal Command Line Setup

Navigate to this directory using your terminal and run:

# 1. Install necessary dependencies
pip install -r requirements.txt

# 2. Run the server
python app.py

Accessing the UI

Open your favorite web browser and go to: 👉 [http://localhost:5000](http://localhost:5000) (Locally) or [Hugging Face Live Demo](https://huggingface.co/spaces/Harsh0809/fruit-quality-classification-cnn) (Cloud)