Object detection in satellite images

gistfile1.txt

Download tiles.
  Zoom 20 is adequate for tennis courts. Zoom 22 is ideal, with diminishing returns.
  C:\Users\Smith\Desktop\dev\cs\QldGlobeApps

Convert tiles to geotiff
  C:\Users\Smith\Desktop\dev\cs\QldGlobeApps
     
Run geotiff through geodeep
  https://github.com/uav4geo/GeoDeep

  pip install -U geodeep

  Detect tennis courts:
    geodeep "E:\data\QldGlobe\2025-04-20 - 4 - QldGlobe exportImage\2025-04-20 - 4 - QldGlobe exportImage (Zoom 20).tiff" aerovision --classes tennis-court -t geojson -o "E:\data\QldGlobe\2025-04-20 - 4 - QldGlobe exportImage\2025-04-20 - 4 - QldGlobe exportImage (Zoom 20).geojson"
    
Post-processing for Tennis Courts
  Remove polygons smaller than tennis courts (typically 260 sq m)
    Layer -> right-click -> Open Attribute Table
    
    Click 'Select features using express'
      $area < 200
      
      Click 'Select Features', Close
      
    Click Pencil to edit
      Click 'Delete selected features'

  To reduce the number of features (to stay within Google My Maps 2000 feature limit)
  
    //Aggregate combines all items into one feature
    Aggregate
      Processing -> Toolbox -> Aggregate
      Remove all items in Aggregates list
      Run
  
    //Dissolve connects neighbouring items (so only somewhat reduces features)
    Dissolve (Note: displays a single confidence score for all items in Google Earth)
      Vector -> Geoprocessing Tools -> Dissolve
      Run
    
Convert geojson to kml:
  Open in QGIS
  
  Layer -> right-click -> Properties
    Choose a style such as 'outline red'
    
  Layer -> right-click -> Export -> Save feature as
    Set 'Format' to KML
    Set 'Symbology export' to 'Symbol Layer Symbology'