🌳 Tree Annotation Tool

A web-based application for automated tree detection and manual annotation using satellite imagery. This tool combines deep learning-based tree detection with interactive manual annotation capabilities to create comprehensive tree datasets.

🚀 Features

• Automated Tree Detection: Uses DeepForest model to automatically detect trees in satellite imagery
• Interactive Annotation: Manual drawing tools for additional tree annotations
• KML/KMZ Support: Upload geographic boundaries via KML/KMZ files
• High-Resolution Satellite Imagery: Downloads high-zoom satellite imagery for detailed analysis
• Cloud Storage Integration: Automatic upload to AWS S3 for image and annotation storage
• Export Capabilities: Save annotations as CSV files with geographic coordinates
• Web-Based Interface: Built with Streamlit for easy access and use

🛠️ Technology Stack

• Frontend: Streamlit, Folium (interactive maps)
• AI/ML: DeepForest (tree detection model)
• Geospatial: GeoPandas, Rasterio, Shapely
• Cloud: AWS S3 (storage)
• Satellite Imagery: TMS/XYZ tile services
• Containerization: Docker

📋 Prerequisites

• Python 3.11+
• AWS S3 bucket and credentials
• Docker (optional, for containerized deployment)

🚀 Quick Start

1. Build the Docker image

   docker build -t annotation-tool .

2. Run the container

   docker run -it --rm \
     -v $(pwd):/app \
     -p 8888:8888 \
     --env-file .env \
     annotation-tool

📖 Usage Guide

1. Upload KML/KMZ File

• Upload a KML or KMZ file containing the geographic boundary of the area you want to analyze
• The tool will automatically download high-resolution satellite imagery for that area

2. Automatic Tree Detection

• The DeepForest model will automatically detect trees in the satellite imagery
• Detected trees are filtered by:
  • Area: ≤ 60 m² (to exclude large objects)
  • Confidence score: ≥ 0.15 (to ensure quality detections)

3. Manual Annotation

• Use the drawing tools on the map to add additional tree annotations
• Draw rectangles around trees that may have been missed by the model
• All manually drawn annotations are labeled as "Tree"

4. Save Annotations

• Click "Save Annotations as CSV" to export all annotations
• The CSV file includes:
  • Geographic coordinates (WGS84)
  • Pixel coordinates
  • Tree areas
  • Confidence scores
  • Source (model vs. manual annotation)
  • Image path reference

5. Download Results

• Download the CSV file locally
• Annotations are also automatically uploaded to your S3 bucket

📁 Project Structure

annotation-tool/
├── app.py                 # Main Streamlit application
├── utils.py              # Utility functions for geospatial operations
├── model-inference.py    # DeepForest model inference utilities
├── requirements.txt      # Python dependencies
├── Dockerfile           # Docker configuration
├── .streamlit/          # Streamlit configuration
│   └── secrets.toml     # AWS credentials (create this)
└── venv/                # Virtual environment (created locally)

🔧 Configuration

Model Configuration

The DeepForest model uses default settings optimized for tree detection:

• Minimum confidence threshold: 0.15
• Maximum tree area: 60 m²
• Satellite imagery zoom level: 19

📊 Output Format

The tool generates CSV files with the following columns:

• geometry: Geographic coordinates (WGS84)
• xmin, ymin, xmax, ymax: Pixel coordinates
• label: Tree classification
• score: Detection confidence (model predictions only)
• area_m2: Tree crown area in square meters
• labelled_by: Source of annotation ("model" or "user")
• image_path: Reference to the satellite image

📝 License

This project is licensed under the MIT License - see the LICENSE file for details.

🙏 Acknowledgments

• DeepForest - Tree detection model
• Streamlit - Web application framework
• GeoPandas - Geospatial data manipulation
• Folium - Interactive maps

🐛 Troubleshooting

Common Issues

1. GDAL Installation Issues

   • On macOS: brew install gdal
   • On Ubuntu: sudo apt-get install libgdal-dev
   • Use conda: conda install gdal

2. AWS Credentials Not Found

   • Ensure .streamlit/secrets.toml exists and contains valid credentials
   • Check AWS IAM permissions for S3 access

3. Memory Issues with Large Images

   • Reduce the area size in your KML file
   • The tool automatically ensures a minimum area of 9,500 m²

4. No Trees Detected

   • Check if the satellite imagery is clear and contains trees
   • Verify the area is not too small or too large
   • Try adjusting the confidence threshold in the code