crop-disease-ai

🌿 AI-Based Crop Disease Detection & Smart Treatment Recommendation System

IEEE Paper Implementation — End-to-end deep learning pipeline for automated crop disease diagnosis with LLM-powered treatment recommendations.

---

Architecture Overview

┌──────────────────────────────────────────────────────────────────┐
│                         FRONTEND                                 │
│     Next.js 14  ·  TypeScript  ·  TailwindCSS                    │
│  ┌──────────┐  ┌──────────┐  ┌────────────┐  ┌──────────────┐   │
│  │  Upload   │→│  Loader  │→│   Result    │  │   History     │   │
│  │ (Dropzone)│  │(Spinner) │  │(Tabs+Badge)│  │ (Pagination)  │   │
│  └──────────┘  └──────────┘  └────────────┘  └──────────────┘   │
└──────────────────────┬───────────────────────────────────────────┘
                       │  HTTP / REST
┌──────────────────────▼───────────────────────────────────────────┐
│                         BACKEND                                  │
│     FastAPI  ·  PyTorch  ·  OpenCV  ·  llama-cpp-python          │
│                                                                  │
│  ┌────────────┐  ┌────────────┐  ┌───────────┐  ┌────────────┐  │
│  │ Preprocess │→│  ResNet18  │→│  Severity  │→│  LLM / Dict  │  │
│  │  (OpenCV)  │  │  (CNN)     │  │ Estimator │  │ (phi-2 GGUF)│  │
│  └────────────┘  └────────────┘  └───────────┘  └─────┬──────┘  │
│                                                        │         │
│  ┌─────────────────────────────────────────────────────▼──────┐  │
│  │                   SQLite (aiosqlite)                        │  │
│  └────────────────────────────────────────────────────────────┘  │
└──────────────────────────────────────────────────────────────────┘

Tech Stack

Supported Diseases (16 Classes)

---

Quick Start

Prerequisites

• Python 3.10+ (Download)
• Node.js 18+ (Download)
• Git
• (Optional) NVIDIA GPU with CUDA for faster inference

---

Backend Setup

Step 1: Navigate to backend

cd crop-disease-ai/backend

Step 2: Create and activate virtual environment

Windows (PowerShell):

python -m venv venv
.\venv\Scripts\Activate.ps1

Windows (CMD):

python -m venv venv
venv\Scripts\activate.bat

Linux / macOS:

python3 -m venv venv
source venv/bin/activate

Step 3: Install dependencies

pip install -r requirements.txt

Note: llama-cpp-python requires a prebuilt wheel on Windows. If it fails to build, install it separately:

pip install llama-cpp-python --prefer-binary --extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cpu

Step 4: Copy environment config

Windows:

copy .env.example .env

Linux / macOS:

cp .env.example .env

Step 5: Download phi-2 LLM model (optional but recommended)

pip install huggingface-hub

Windows (PowerShell):

cd models
python -c "from huggingface_hub import hf_hub_download; hf_hub_download(repo_id='TheBloke/phi-2-GGUF', filename='phi-2.Q4_K_M.gguf', local_dir='.')"
cd ..

Linux / macOS:

cd models
python -c "from huggingface_hub import hf_hub_download; hf_hub_download(repo_id='TheBloke/phi-2-GGUF', filename='phi-2.Q4_K_M.gguf', local_dir='.')"
cd ..

The file is ~1.8 GB. Without it, the system uses a built-in expert fallback dictionary and works fully without any LLM.

Step 6: Train the CNN model (optional for demo)

# Download PlantVillage dataset and extract to a folder
# Expected structure: dataset_root/ClassName/image.jpg

python train.py --data-dir /path/to/plantvillage --epochs 25 --batch-size 32

This saves:

• models/disease_model_scripted.pt -- TorchScript model
• models/disease_model_best.pth -- Checkpoint
• models/class_mapping.json -- Class index mapping
• models/training_history.json -- Loss/accuracy curves

Without training, the system runs in demo mode using an ImageNet-pretrained backbone. Predictions will vary per image but won't be accurate.

Step 7: Start the backend server

python run.py

You should see:

  Host:   0.0.0.0
  Port:   8000
  Device: cpu
  LLM:    gguf (or fallback)

• API server: http://localhost:8000
• Swagger docs: http://localhost:8000/docs
• Health check: http://localhost:8000/api/health

---

Frontend Setup

Open a new terminal (keep the backend running).

Step 1: Navigate to frontend

cd crop-disease-ai/frontend

Step 2: Install Node.js dependencies

npm install

Step 3: Start development server

npm run dev

You should see:

  Local: http://localhost:3000

Step 4: Open in browser

• Home (Diagnose): http://localhost:3000
• History: http://localhost:3000/history

---

Verify Everything Works

1. Open http://localhost:3000 in your browser
2. Drag and drop (or click to select) a leaf image (JPEG/PNG, max 10MB)
3. Click "Analyze Disease"
4. View disease prediction, confidence score, severity badge, and AI recommendations
5. Check http://localhost:3000/history for past predictions

Common Issues

---

API Endpoints

POST /api/predict

Request: multipart/form-data

• file (required): JPEG/PNG image
• region (optional): Geographic region for localized advice

Response:

{
  "disease": "Tomato Late Blight",
  "confidence": 0.943,
  "severity": "severe",
  "recommendation": {
    "description": "Late blight is caused by...",
    "causes": "Phytophthora infestans...",
    "treatment": "Apply copper-based fungicide...",
    "prevention": "Use resistant varieties..."
  },
  "timestamp": "2024-01-15T10:30:00",
  "filename": "leaf_sample.jpg"
}

---

Project Structure

crop-disease-ai/
├── backend/
│   ├── app/
│   │   ├── __init__.py
│   │   ├── config.py          # Settings & disease classes
│   │   ├── preprocessing.py   # OpenCV image pipeline
│   │   ├── model.py           # ResNet18 classifier
│   │   ├── severity.py        # Severity estimation
│   │   ├── llm.py             # LLM engine + fallback dict
│   │   ├── database.py        # Async SQLite operations
│   │   ├── schemas.py         # Pydantic response models
│   │   └── main.py            # FastAPI app & endpoints
│   ├── context/
│   │   ├── SYSTEM_ARCHITECTURE.md
│   │   ├── MODEL_DETAILS.md
│   │   ├── FUTURE_WORK.md
│   │   └── DEPLOYMENT_GUIDE.md
│   ├── models/                # Model weights directory
│   ├── train.py               # Training script
│   ├── run.py                 # Uvicorn entry point
│   ├── requirements.txt
│   └── .env.example
├── frontend/
│   ├── app/
│   │   ├── components/
│   │   │   ├── Upload.tsx     # Drag-and-drop upload
│   │   │   ├── Result.tsx     # Prediction results + tabs
│   │   │   ├── Loader.tsx     # Loading animation
│   │   │   └── SeverityBadge.tsx
│   │   ├── history/
│   │   │   └── page.tsx       # Prediction history
│   │   ├── layout.tsx         # Root layout
│   │   ├── page.tsx           # Home / diagnosis page
│   │   └── globals.css
│   ├── lib/
│   │   └── api.ts             # Axios API client & types
│   ├── package.json
│   ├── tailwind.config.ts
│   └── tsconfig.json
└── README.md

---

LLM Configuration

The system supports two LLM backends (configurable via LLM_BACKEND in .env):

Set LLM_BACKEND=none to disable the LLM entirely and rely on the built-in expert dictionary.

---

Training Details

• Architecture: ResNet18 with custom fully-connected head (512 → 16)
• Strategy: Freeze backbone for 5 epochs → unfreeze all layers
• Optimizer: Adam (lr=1e-4, weight_decay=1e-5)
• Scheduler: ReduceLROnPlateau (factor=0.5, patience=3)
• Early Stopping: Patience 5, monitors validation loss
• Augmentation: RandomResizedCrop, HorizontalFlip, Rotation(±15°), ColorJitter
• Export: TorchScript (.pt) for production inference

---

Future Work

1. Multi-Dataset Fusion — Combine PlantVillage with FGVC, iNaturalist datasets
2. Severity-Based Dosage — Calibrate treatment quantities to severity level
3. Edge Deployment — ONNX / TFLite / CoreML export for mobile
4. Regional Personalization — Weather API + location-aware recommendations
5. Prompt Logging & RAG — Retrieval-augmented generation for evolving advice
6. Modular Crop Expansion — Plug-in architecture for new crops/diseases
7. Farmer Feedback Loop — Collect user corrections to improve predictions

See backend/context/FUTURE_WORK.md for detailed plans.

---

References

• IEEE Paper: "AI-Based Crop Disease Detection and Smart Treatment Recommendation System"
• PlantVillage Dataset: Hughes & Salathé (2015)
• ResNet: He et al., "Deep Residual Learning for Image Recognition" (2015)
• phi-2: Microsoft Research (2023)

---

License

MIT License — See LICENSE for details.