🌱 Plant Disease Detector

A deep learning project that classifies potato leaf images as Healthy, Early Blight, or Late Blight using a Convolutional Neural Network (CNN), served through a FastAPI backend and a simple web frontend.

Overview

The project has three parts:

1. Training — a Jupyter notebook that trains a CNN on the PlantVillage potato leaf dataset.
2. API — a FastAPI server that loads the trained model and exposes a /predict endpoint for image classification.
3. PlantCare — a lightweight HTML/CSS/JS frontend where a user uploads a leaf image and sees the predicted disease class and confidence score.

Project Structure

Parent Folder/                 # choose your desired folder name and modify the path names accordingly
├── api/
│   └── main.py                # FastAPI app serving predictions
├── models/
│   └── model_1.keras          # Trained CNN model
├── Training/
│   ├── model-training.ipynb   # Notebook used to train the model
│   └── PlantVillage/          # Training dataset (Early Blight / Late Blight / Healthy)
├── PlantCare/
│   ├── index.html             # Upload UI
│   ├── app.js                 # Calls the API and renders results
│   └── style.css              # Styling
└── README.md

Model

• Architecture: Sequential CNN — resizing/rescaling and data augmentation (random flip + rotation) followed by 6 Conv2D + MaxPooling blocks, then Dense layers with a softmax output.
• Input size: 256 × 256 RGB images
• Classes: Early Blight, Late Blight, Healthy
• Training split: 80% train / 10% validation / 10% test
• Epochs: 50, batch size 32
• Dataset: ~2,150 potato leaf images across the three classes (PlantVillage subset)

The trained model is saved at models/model_1.keras.

Getting Started

1. Requirements

• Python 3.9+
• Node not required — the frontend is plain HTML/JS

Install the Python dependencies:

pip install fastapi uvicorn tensorflow numpy pillow python-multipart

2. Run the API

api/main.py currently loads the model from a hardcoded Windows path:

MODEL = tf.keras.models.load_model(r"C:\AI ML Potato Project\models\model_1.keras")

Update this to a relative path before running elsewhere, e.g.:

MODEL = tf.keras.models.load_model("../models/model_1.keras")

Then start the server from the api directory:

cd api
python main.py

The API will run at http://localhost:8000. Check it's alive:

curl http://localhost:8000/ping

3. Use the Frontend

Open PlantCare/index.html in a browser (or serve the folder with a simple HTTP server). Upload a potato leaf image and click Upload & Predict — the app calls the API's /predict endpoint and displays the predicted class and confidence percentage.

Note: app.js calls the API at http://127.0.0.1:8000/predict/ — make sure the API is running locally on port 8000, or update this URL to match your deployment.

4. Retrain the Model (optional)

Open Training/model-training.ipynb in Jupyter, point it at the PlantVillage dataset folder, and run all cells. The final cell saves the new model to ../models/model_{version}.keras.

API Reference

Method	Endpoint	Description
GET	/ping	Health check
POST	/predict	Accepts an image file, returns predicted class and confidence

Example response:

{
  "class": "Late Blight",
  "confidence": 0.97
}

Author

Kamran Nazir — Islamic University of Science & Technology (IUST), Kashmir

Acknowledgements

The development of this project was guided by the following Potato Disease Classification from "Code Basics", with modifications and adaptations made to suit the requirements of this project.