Krishi AI

Krishi AI is a plant disease diagnosis project with a Flutter mobile app and a FastAPI backend powered by a PyTorch model. The app lets you pick or capture a leaf photo, send it to the backend, and show the prediction result plus an optional Grad-CAM heatmap.

Tech Stack

• Flutter for the mobile app
• Dart for the client logic
• FastAPI for the backend API
• PyTorch and TorchVision for inference
• OpenCV and Grad-CAM for explainability output
• PostgreSQL for diagnosis history and feedback
• Docker for backend deployment

Project Structure

• app/ Flutter application
• backend/ FastAPI server and model inference code
• model/ local PyTorch model weights
• data/ raw and verified image storage

Features

• Capture a leaf photo from camera or gallery
• Upload the image to the backend for diagnosis
• Show disease name, confidence, and AI heatmap
• Submit feedback to improve future predictions
• Store diagnosis history in PostgreSQL

Requirements

• Flutter 3.x or newer
• Python 3.12 or newer
• PostgreSQL database
• A trained .pth model file in model/
• Backend environment variables configured in backend/.env
• App environment variables configured in app/.env

Backend Setup

Install dependencies:

cd backend
pip install -r requirements.txt

Create a backend/.env file with values like:

DATABASE_URL=postgresql://user:password@localhost:5432/krishi_ai
LOCAL_MODEL_PATH=model/best_plant_model.pth
TRAINING_API_KEY=your_training_api_key
NUM_CLASSES=38
SKIP_HEATMAP=false
PROJECT_NAME=Krishi AI Backend

Run the API locally:

cd backend
uvicorn main:app --host 0.0.0.0 --port 7860 --reload

The API will be available at:

• http://localhost:7860/health
• http://localhost:7860/api/diagnose

Flutter App Setup

Install packages:

cd app
flutter pub get

Create an app/.env file and point it to your backend URL:

BASE_URL=http://10.16.25.26:5407

Run the app:

cd app
flutter run

If you are testing on a physical device, make sure the backend URL is reachable from that device and update it from the app settings screen if needed.

How to Use

1. Start the backend server.
2. Open the Flutter app.
3. Tap camera or gallery to choose a plant leaf image.
4. Tap Diagnose Plant.
5. Review the prediction, confidence score, and heatmap.
6. If the result is wrong, submit feedback from the app.

Deployment

Backend Deployment

The backend includes a Dockerfile, so you can build and run it with Docker:

cd backend
docker build -t krishi-ai-backend .
docker run -p 7860:7860 --env-file .env krishi-ai-backend

Before deploying, make sure the following are available in the container or host environment:

• DATABASE_URL
• TRAINING_API_KEY
• NUM_CLASSES
• LOCAL_MODEL_PATH

Flutter Deployment

Build the Android release APK:

cd app
flutter build apk --release

If you want iOS or web builds, use the normal Flutter release commands for those targets.

License

This project is licensed under the MIT License. See LICENSE for the full text.

Notes

• The Flutter app loads its backend URL from app/.env and also lets you update it from the settings dialog.
• The backend expects image uploads as multipart form data with the file field.
• The model must be loaded before diagnosis will work; if it is missing, the backend returns a helpful error.