TreeDetection

A Python tool for processing large GeoTIFF images by slicing them into smaller tiles, running YOLO object detection, and generating both a stitched output image and a shapefile of detections.

Overview

This tool is designed to handle large GeoTIFF images by breaking them down into manageable tiles, running YOLO object detection on each tile, and then reassembling them into a complete image. It also generates a shapefile containing the geographic coordinates of all detections. The tool can be used either through a command-line interface or a web interface.

Features

• Slice large GeoTIFF images into smaller tiles
• Run YOLO object detection on each tile
• Draw detection boxes on the processed images
• Generate a shapefile of detection coordinates
• Preserve original GeoTIFF metadata and coordinate system
• Web interface for easy file upload and processing
• Command-line interface for batch processing

Requirements

• Python 3.10
• Required packages:
  • numpy
  • PIL (Python Imaging Library)
  • rasterio
  • fiona
  • shapely
  • ultralytics (YOLO)
  • flask (for web interface)

Installation

1. Clone the repository:

   git clone https://github.com/LJMUAstroecology/TreeDetection.git
   cd TreeDetection

Using Conda

1. Create a new conda environment:

   conda create -n TreeDetection python=3.10

2. Activate the environment:

   conda activate TreeDetection

3. Install the required packages:

   pip install -r requirements.txt

Usage

Web Interface

1. Start the web server:

   python webapp.py

2. Open your web browser and navigate to http://localhost:5000

3. Upload a GeoTIFF file through the web interface

4. The tool will process the image and provide download links for:

   • The stitched GeoTIFF with detection boxes
   • The shapefile containing detection coordinates

Command Line Interface

The main script EndToEnd.py provides a complete pipeline for image processing:

from EndToEnd import run_end_to_end

run_end_to_end(
    original_tiff_path='path/to/input.tif',
    sliced_dir='sliced_tiles',
    detected_dir='detected_tiles',
    output_tiff_path='outputs/stitched_output.tif',
    output_shp_path='outputs/detections.shp',
    tile_size=(512, 512),  # Optional: customize tile size
    model_path='best.pt',  # Optional: specify YOLO model
    conf=0.3  # Optional: detection confidence threshold
)

File Structure

• EndToEnd.py: Main processing script
• webapp.py: Web interface implementation
• best.pt: YOLO model weights
• uploads/: Directory for uploaded files
• outputs/: Directory for final output files
• sliced_tiles/: Directory for intermediate sliced tiles
• detected_tiles/: Directory for processed tiles with detections

Output Files

1. Stitched GeoTIFF: Contains the original image with detection boxes drawn on it

   • Preserves all original GeoTIFF metadata
   • Maintains the original coordinate system
   • Detection boxes are drawn in red

2. Shapefile: Contains the geographic coordinates of all detections

   • Includes .shp, .shx, .dbf, and .prj files
   • Uses the same coordinate system as the input GeoTIFF
   • Each detection is stored as a polygon feature

Notes

• The tool uses YOLO for object detection, which requires a trained model (provided as best.pt)
• Make sure to have sufficient disk space for the intermediate files
• The web interface is intended for development/testing; for production use, consider adding proper security measures
• The shapefile output requires all associated files (.shp, .shx, .dbf, .prj) to be used properly in GIS software

License: TBD

Add a funny joke below here