An operational framework for object-based land use classification of heterogeneous rural landscapes

Research output: Research - peer-reviewJournal article

DOI

  • Gary Richard Watmough
  • Cheryl Palm
    Cheryl PalmColumbia University's Earth InstituteUnited States
  • Clare Sullivan
    Clare SullivanColumbia University's Earth InstituteUnited States
The characteristics of very high resolution (VHR) satellite data are encouraging development agencies to investigate its use in monitoring and evaluation programmes. VHR data pose challenges for land use classification of heterogeneous rural landscapes as it is not possible to develop generalised and transferable land use classification definitions and algorithms. We present an operational framework for classifying VHR satellite data in heterogeneous rural landscapes using an object-based and random forest classifier. The framework overcomes the challenges of classifying VHR data in anthropogenic landscapes. It does this by using an image stack of RGB-NIR, Normalised Difference Vegetation Index (NDVI) and textural bands in a two-phase object-based classification. The framework can be applied to data acquired by different sensors, with different view and illumination geometries, at different times of the year. Even with these complex input data the framework can produce classification results that are comparable across time. Here we describe the framework and present an example of its application using data from QuickBird (2 images) and GeoEye (1 image) sensors.
Original languageEnglish
JournalInternational Journal of Applied Earth Observation and Geoinformation
Volume54
Pages (from-to)134-144
Number of pages11
ISSN0303-2434
DOIs
StatePublished - 1 Feb 2017

See relations at Aarhus University Citationformats

ID: 103246940