🌱 AgriSense-X: Explainable Multimodal Crop Disease Detection & Advisory System

An AI-powered agricultural assistant combining Computer Vision and Large Language Models to provide intelligent crop disease diagnosis with actionable treatment advice.

🎯 Project Highlights

What Makes This Special?

1. Explainable AI - Not just predictions, but visual explanations using GradCAM showing exactly where the AI is looking
2. Conversational Expert - Chat with Claude AI for follow-up questions about treatments, costs, and concerns
3. Multilingual Support - Advice in English, Hindi and Nepali for broader farmer accessibility
4. Treatment Timeline - Actionable checklists and progress tracking
5. Low-Confidence Alerts - Differential diagnosis for uncertain predictions

🏗️ Architecture

┌─────────────────────────────────────────────────────────────┐
│                    AgriSense-X System                        │
├─────────────────────────────────────────────────────────────┤
│                                                               │
│  ┌───────────────┐    ┌──────────────┐    ┌──────────────┐ │
│  │   Image       │───▶│  MobileNetV2 │───▶│  Disease     │ │
│  │   Upload      │    │  Classifier  │    │  Prediction  │ │
│  └───────────────┘    └──────────────┘    └──────────────┘ │
│                              │                      │         │
│                              ▼                      ▼         │
│                       ┌──────────────┐    ┌──────────────┐  │
│                       │   GradCAM    │    │  Claude AI   │  │
│                       │ Explainability│   │   Advisory   │  │
│                       └──────────────┘    └──────────────┘  │
│                              │                      │         │
│                              ▼                      ▼         │
│                       ┌─────────────────────────────────┐   │
│                       │    Streamlit Web Interface      │   │
│                       │  • Visual Explanations          │   │
│                       │  • Conversational Chat          │   │
│                       │  • Multilingual Support         │   │
│                       │  • Analytics Dashboard          │   │
│                       └─────────────────────────────────┘   │
└─────────────────────────────────────────────────────────────┘

📁 Project Structure

AgriSense-X/
├── model/
│   ├── train.py             
│   ├── predict.py           
│   ├── gradcam.py           
│   └── model.pth             
│
├── llm/
│   └── advisor.py            
│                             
│
├── data/
│   └── plantvillage/         
│       ├── Tomato___Late_blight/
│       ├── Potato___Early_blight/
│       └── ...
│
├── app.py                   
├── requirements.txt         
└── README.md                

🚀 Setup Instructions

1. Install Dependencies

pip install -r requirements.txt

2. Configure Claude API

Get your API key from: https://console.anthropic.com/

Add to llm/advisor.py:

client = anthropic.Anthropic(api_key="your-api-key-here")

Or set as environment variable:

export ANTHROPIC_API_KEY="your-api-key-here"

3. Create model

To save the model run:

model/train.py

4. Run the Application

streamlit run app.py

The app will open at http://localhost:8501

📊 Technical Metrics

• Model: MobileNetV2 (efficient, deployable)

• Training: PlantVillage dataset (15 disease classes)

• Accuracy:

  Epoch 1 Loss: 0.2485
  Epoch 2 Loss: 0.0417
  Epoch 3 Loss: 0.0258
  Epoch 4 Loss: 0.0142
  Epoch 5 Loss: 0.0169
  

  Loss decreased by 93% from epoch 1 to 5, showing excellent convergence. The slight increase in epoch 5 suggests we could benefit from early stopping or learning rate scheduling for deployment.

• Explainability: GradCAM attention visualization

• Languages: 4 (English, Hindi, Nepali)

• Advisory: Context-aware using Claude Sonnet 4

📚 References for Your Report

• PlantVillage Dataset: https://www.kaggle.com/datasets/emmarex/plantdisease/data
• GradCAM Paper: https://arxiv.org/abs/1610.02391
• MobileNetV2: https://arxiv.org/abs/1801.04381
• Claude Technical Details: https://www.anthropic.com/claude
• Agricultural Disease Impact: FAO Reports on Crop Loss

🎬 Demo Video