Rice Disease Detection

A web application that detects rice leaf diseases using a YOLOv5 model (TensorFlow Lite). Upload an image and get the top-3 detections with bounding boxes drawn on the image.

Detected Diseases

Label	Description
Blight	Bacterial leaf blight
Brown spot	Brown spot disease
False Smut	False smut disease
Healthy	Healthy leaf
Leaf Smut	Leaf smut disease
Rice blast	Rice blast disease
Stem Rot	Stem rot disease
Tungro	Tungro disease

Setup

npm install
npm start

Open http://localhost:3000.

Deploy to Vercel

npm i -g vercel
vercel --prod

The project includes vercel.json and api/index.js for serverless deployment. The model loads lazily on the first request and is cached for subsequent calls.

API

POST /predict — Upload an image and receive detections.

• Content-Type: multipart/form-data
• Field: image (file)
• Response:

  {
    "detections": [
      {
        "class": 4,
        "label": "Leaf Smut",
        "confidence": 0.43,
        "x1": 121.01,
        "y1": 80.63,
        "x2": 367.65,
        "y2": 440.11
      }
    ],
    "numDetections": 3,
    "originalSize": { "width": 1920, "height": 1080 }
  }

Coordinates are in the original image's pixel space. Only the top-3 highest-confidence detections are returned.

Project Structure

├── api/index.js          # Vercel serverless entry point
├── asset/
│   ├── best-fp16.tflite  # YOLOv5 TFLite model
│   └── labels.txt        # Class labels
├── public/
│   └── index.html        # Frontend UI
├── index.js              # Express server & model inference
├── package.json
└── vercel.json           # Vercel deployment config