🌿 Crop Doctor
AI-powered crop disease detection platform for Indian farmers — a 3-model system combining leaf image diagnosis, weather-based disease risk prediction, and ICAR-aligned treatment advisory.
📱 App Screens Home Dashboard (crop selection + live weather) Scan Leaf (camera / gallery upload) Disease Result (confidence + severity + treatment + risk)
📌 Table of Contents Overview The Problem What It Does Architecture Models Datasets Project Structure Setup & Installation Running the Backend Running the Flutter App API Reference Model Performance What Makes This Different Roadmap Team
🚀 Overview
Crop Doctor is an end-to-end AI platform that helps Indian farmers detect crop diseases early and take action before yield is lost.
It combines three AI models into a single mobile app:
Model 1 — Diagnoses disease from a leaf image Model 2 — Provides ICAR-aligned treatment recommendations Model 3 — Predicts disease risk using live weather data
✅ Fully working Android APK ✅ Tested on real device ✅ Real-time image + weather predictions
India loses ~35% of annual crop yield due to plant diseases (~₹2,400 crore/year).
Key challenges:
58% farmers lack timely expert advice 2–3 day detection delay → disease spreads to 40–60% of field Existing apps fail in real-world conditions (trained on lab images) No preventive, weather-based prediction tools No ICAR-aligned localized treatment guidance 🌾 What It Does Farmer Workflow Select crop (Rice, Tomato, Wheat, etc.) View live weather + disease risk Upload leaf image Get diagnosis + severity + treatment View upcoming disease risks Output per Scan Disease name (86 classes) Confidence score Severity: Low / Moderate / High Affected area % Chemical treatment + dosage Organic alternative Prevention steps Weather-based risk forecast
🏗️ Architecture
Flutter App (Home · Scan · Result Screens) │ ▼ FastAPI Backend ├── /predict/image ├── /predict/weather-risk └── /advisory │ ├── Model 1 → Image Classifier (ONNX) ├── Model 2 → Advisory Engine (ICAR KB) └── Model 3 → Weather Risk (XGBoost)
🤖 Models Model 1 — Leaf Image Classifier Architecture: MobileNetV3 (supports EfficientNet-B3, ResNet-50) Format: ONNX Runtime Classes: 86 Input: 224×224 Inference: Test-Time Augmentation (TTA) Model 2 — Treatment Advisory JSON-based ICAR knowledge base Provides: Chemical treatment Organic alternatives Dosage & frequency Prevention steps Optional LLM fallback Model 3 — Weather Risk Predictor Algorithm: XGBoost Accuracy: 96.53% Data: NASA POWER (historical) OpenWeatherMap (live) Predicts disease risk before symptoms appear
📊 Datasets Dataset Images Role PlantVillage 54,306 Base training PlantDoc 2,598 Field realism MultiCrop Tamil Nadu 23,000+ Local ground truth NASA POWER — Weather training
Final dataset: 35,275 images · 86 classes
🌍 Supported Crops Tomato Rice Wheat Cotton Maize Groundnut Potato Chilli Sugarcane Soybean
🏙️ Cities Covered
Chennai · Bengaluru · Hyderabad · Mumbai · Pune · Ahmedabad · Jaipur · Lucknow · Bhopal · Kolkata
📁 Project Structure crop_disease_platform/
backend/ ├── models/ ├── src/ │ ├── model1/ │ ├── model2/ │ ├── model3/ │ ├── advisory/ │ └── api/ └── data/
flutter_app/ └── lib/ ├── screens/ └── services/
⚙️ Setup & Installation Prerequisites Python 3.10+ Flutter 3.x OpenWeatherMap API key
- Clone Repo git clone https://github.com/YOUR_USERNAME/crop-doctor.git cd crop-doctor
- Backend Setup cd backend python -m venv venv
venv\Scripts\activate # Windows source venv/bin/activate # Mac/Linux
pip install -r requirements.txt 3. Environment Variables
Create .env:
OPENWEATHER_API_KEY=your_key MODEL2_ENDPOINT=
Swagger Docs: 👉 http://localhost:8000/docs
📱 Running Flutter App cd flutter_app flutter pub get flutter run
Update API URL:
static const String _baseUrl = 'http://192.168.1.XXX:8000';
🔌 API Reference POST /predict/image
Returns:
Disease Confidence Severity Treatment Top predictions POST /predict/weather-risk
Returns:
Weather summary Disease risk Prevention tips POST /advisory
Returns:
ICAR-based treatment plan
📈 Model Performance Model 1 Accuracy: 48.4% (baseline) Macro F1: 41.8% Model 3 Accuracy: 96.53% Precision: 94.42% Recall: 97.33% ROC AUC: 0.9952
🌟 What Makes This Different Feature Crop Doctor Typical Apps Data Real + Indian Lab only Prediction ✔ Weather-based ✖ No Treatment ✔ ICAR aligned ✖ Generic Deployment ✔ Mobile app ✖ Demo only Models 3 combined 1
🛣️ Roadmap EfficientNet-B3 upgrade Field evaluation improvements Multi-language (Tamil, Hindi) Cloud deployment (AWS / Railway) Offline inference (ONNX mobile) UAV / drone integration
👥 Team
Model Evaluation & Optimization Ensemble Learning Architect Data Science & Feature Engineering
📌 Summary
Crop Doctor is not just a detection app — it is a complete AI-driven decision support system for farmers, combining:
Diagnosis Treatment Prediction into one unified platform.