SMGIS Universal GIS Tools

Hybrid LULC Classification

Automatic Land Use Land Cover classification — K-means clustering labeled with NDVI/NDWI/NDBI — refined with your own samples if you want. No training-data pipeline required to get a real result.

Area of Interest (optional)
Restrict processing to a boundary you provide — leave this blank to process the entire image.
Area of Interest set
Drag & Drop or Click to Browse
Multi-band imagery (GeoTIFF, ERDAS IMAGINE, or JPEG2000) — Red, Green, and NIR bands required
Max file size: 200 MB

Choose how you want to classify this image

Click on the image to add a sample.

What do you want delivered?

How Hybrid Classification Works

Automatic Baseline

K-means clustering groups pixels by spectral similarity, then each cluster gets a real name (Water, Vegetation, Built-up, Bare Soil) from its actual NDVI/NDWI/NDBI values — not guessed, computed.

Refine or Extend

Add your own samples to strengthen an automatic class, or define an entirely new one — both use the exact same underlying signature method.

Outlier Protection

Every new sample is checked against the automatic baseline — if it looks like it belongs to a different class, you get a warning before it's used, not a silently wrong result.

Clean Vector Output

Tiny, isolated misclassified pixels are filtered out before vectorizing — real features survive, noise doesn't become a pile of tiny polygons.

Frequently Asked Questions

What does "Hybrid" mean here?

Every class — whether automatically generated or built from your own clicked samples — is represented the same way, as a spectral signature. The automatic path is genuinely unsupervised (no training data); refining a class or adding your own is genuinely supervised. This tool supports both, and you choose how much of each to use.

What image bands do I need?

Red, Green, and Near-Infrared (NIR) at minimum, for the NDVI/NDWI calculations this tool relies on. A SWIR band (common in Landsat-style 6-band imagery) enables a real built-up/bare-soil distinction via NDBI — without it, that distinction falls back to a coarser brightness-based estimate.

How many sample clicks does adding a class take?

6 initial samples minimum, then up to 6 spatially-spread candidate pixels per verification round to quickly confirm or reject — never a long list, and always optional after the first round.

How accurate is this?

Genuinely useful for a fast, real classification — but this is unsupervised classification refined by your samples, not survey-grade without real ground-truthing. Two spectrally similar but semantically different surfaces can still be confused.

What outputs can I export?

Choose any combination of a classified raster (GeoTIFF), labeled vector polygons, and a branded PDF report with category-wise area — delivered together in one download.

Need a validated, ground-truthed LULC classification for a real project? Our team can build that with proper accuracy assessment.

Request a Consultation