Detecting tree canopies with Unsupervised Classification in SEXTANTE

The aim of this proof of concept model is to automatically classify tree canopies from aerial images. An aerial image showing a small area (about  8km by 5km) has been used for this.

  1. The first step is to perform unsupervised classification. For this the Unsupervised KMeans image classification module from OTB has been used. We’ve chosen to create 25 classes and set training size to 20. The results looks like:Classification
  2. The next step is to reclass the output image, so only classes representing tree canopies are retained. GRASS r.reclass or SAGA Reclassify Grid Values can be used
  3. Accuracy of the classification can be improved. Small areas like mall bushes, shrubs etc. can be removed with GRASS r.reclass.area.greaterSmall clusters removed
  4.   Another problem are the water bodies . To deal with this problem GRASS r.reclass.area.lesser  is appliedSmall & big clusters removed
  5.     Next step is to clip the resulting layer with the Corine Land Cover dataCLC difference