Intelligent Crop Health Advisor

AI system to identify crop type, diagnose plant diseases from images, and deliver multilingual text/voice advice with offline-first deployment.

Highlights

• Crop and disease identification via separate CNNs for precision and speed.
• Actionable advice from a local, fine-tuned LLM with safety filters.
• Multilingual chatbot with STT/TTS for low-connectivity, accent-robust use.
• Sustainability focus: targeted treatments to reduce pesticide overuse.

Tech Stack

• DL: PyTorch/TensorFlow, OpenCV, PIL, NumPy, Pandas
• LLM/Runtime: Transformers, Optimum, ONNX Runtime, TensorRT
• Backend: FastAPI or Node.js, PostgreSQL
• Voice: Whisper (STT), VITS (TTS)

Datasets

• PlantVillage (~54k images)
• PlantDoc (~2.6k images)
• Custom regional crop/disease images
• Common Voice (selected languages)

Models

• Crop ID: MobileNetV2 (transfer learning)
• Disease: ResNet50/EfficientNet (class-balanced)
• Advice: Gemma 3 (via Ollama)
• Chatbot: Whisper + VITS

Performance Targets

• Crop ID: ≥90% accuracy, ≤1s inference
• Disease: ≥85% F1, high early-stage recall
• Advice: ≤2s response, safe/relevant outputs
• Chatbot: ≥85% intent accuracy, tracked WER

Quickstart

Prerequisites

• Python 3.10+
• CUDA (recommended for faster inference)
• ffmpeg (for voice features)
• PostgreSQL (for data storage)
• Ollama (for running LLM locally)

Setup

1. Clone the repository and install requirements:

   git clone <repository-url>
   cd Crop-Diseases-ChatBot-Advice
   python -m venv venv
   source venv/Scripts/activate 
   pip install -r requirements.txt

2. Install Ollama:

   • Download and install from Ollama's official website
   • Or use the following command:

     curl -fsSL https://ollama.com/install.sh | sh

3. Pull the required LLM model:

   ollama pull gemma3

4. Start the Ollama server (if not running automatically):

   ollama serve

5. Start the application:

   python app.py

   The application will be available at http://127.0.0.1:7860

Usage

1. Upload a crop image for analysis
2. View the diagnosis and care advice
3. Ask follow-up questions in the chat interface

For development with the API:

• POST image to /predict/crop for crop identification
• POST to /predict/disease for disease diagnosis
• Request advice via /advice
• Chat via /chat (supports text/audio)

API (minimal)

• POST /predict/crop → {crop, top_k, confidence}
• POST /predict/disease → {disease, severity, notes}
• POST /advice → {treatment, prevention, followup}
• POST /chat → {text_out, audio_out?}

Project Structure

• models/ (cnn/, llm/, stt_tts/)
• api/ (routes/, schemas/)
• data/ (raw/, processed/)
• scripts/ (train_, export_, serve_*)
• configs/ (.env, model/config.yaml)

Roadmap

• Expand regional datasets and balance rare diseases
• Improve early-stage recall and edge latency
• Broaden language/accent coverage
• Add offline bundles and edge profiles

License

Choose a permissive OSS license; respect dataset licenses.