Skip to content

Landuse classification

Problem definition

  • We have the following datasets:
    • An existing Landuse vector dataset
    • Orthorectified Aerial images (Digital Ortho Photos = DOP)
    • Satellite images (SAT), e.g. Sentinel 2 10m
  • We want to select / extract landcover classes from landuse classes.
  • We want to use Support Vector Machines (or other Machine Learning Classifiers) for classifying landcover classes.
  • We want to support Multiple classes using a One versus All (OvA) strategy or One vs. Rest.
  • We want to parallelize the computation in our Spark cluster using Geotrellis for data loading and export.


  • Describe the idea
  • Maybe add some geotrellis examples

Classification of Aerial Images according to Land Use Classes (using Land Cover Classes as intermediate)


  • Machine Learning
  • Training: Multiclass SVM
  • Geotrellis

Scala code snippets

type SpatialMultibandRDD = RDD[(SpatialKey, MultibandTile)] with Metadata[TileLayerMetadata[SpatialKey]]

// reading from Hadoop Layer (HDFS)
val rdd : SpatialMultibandRDD = biggis.landuse.api.readRddFromLayer(LayerId(layerName, zoom))

// writing to Hadoop Layer (HDFS)
biggis.landuse.api.writeRddToLayer(rdd, LayerId(layerName, zoom))


  • Classification of Aerial Images May-Aug 2016
  • Layerstacking: Aerial Images + Satellite images (IR, Resolution 2m)
  • Training of a Multiclass SVM with manually selected training data (classified image tiles)

Further Steps

  • Adding additional Layers, e.g.
  • Terrain Height
  • Homogeneity of Texture
  • Using Other Classififiers