raster2poly

Classify rasters and vectorise the result to clean polygons — in three lines of code.

Supports unsupervised clustering (KMeans), supervised classification (Random Forest from ROI shapefiles), and rule-based DN thresholds. Outputs are dissolved, filtered GeoDataFrames ready for GIS.

Installation

pip install raster2poly

Quick start

Unsupervised (KMeans)

from raster2poly import RasterClassifier

clf = RasterClassifier("satellite_image.tif")
gdf = clf.unsupervised(n_clusters=6, algorithm="mini_batch_kmeans")
clf.save(gdf, "classes.gpkg")

Supervised (ROI shapefile)

gdf = clf.supervised("training_rois.shp", class_col="class_id")
clf.save(gdf, "supervised.shp")

The ROI file can contain Points or Polygons (or both). For polygons, every pixel inside the geometry is used as a training sample — far more robust than a single zonal mean.

Rule-based (DN ranges)

rules = {
    1: [(4, 0.15, 1.0), (5, 0.0, 0.10)],  # high Red, low NIR → built-up
    2: [(5, 0.25, 1.0)],                     # high NIR → vegetation
}
gdf = clf.from_dn_ranges(rules)

Band numbers are 1-based. A pixel must satisfy all conditions in the list to be assigned that class.

Key improvements over the original script

Output format

The returned GeoDataFrame has two columns:

• class_id (int) — the class label
• geometry — dissolved polygons

Save to any format: .shp, .gpkg, .geojson.

License

MIT