Extracting Nature Areas using Object-oriented Analysis

Laboratory of Geo-Information Science and Remote Sensing
  Education
  Research
  Publications
  Models
  News & Calendar
  News
  Calendar
  Archive
  News
  Calendar
  2011
  2010
  2009
  2008
  2007
  2006
  2005
  2012
  Staff
  Equipment
  Contact details
  Workshops

23 Apr 2009 09:30 - 23 Apr 2009 10:00
Unit: Laboratory of Geo-Information Science and Remote Sensing
Location: Gaia 2
Organisation: Wageningen University

By Lalitya Narieswari

Abstract
  
A so called non-intervention management scheme has been adopted in a large part of the nature area National Park Veluwezoom near Rheden in the Netherlands. This means that natural processes like storm, grazing, climate, succession and diseases are allowed to occur and become a dominant factor that changes the structure and ecosystem in the area. There is a need to monitor how the applied management influences the environment and the habitat of the nature area.

This study focuses on developing an approach to extract the different types of land cover which characterize the nature areas in the study area based on object oriented analysis using aerial photographs. Three subset areas were chosen and considered as representative areas. The study aimed to create a rule set transferable to other datasets to enable automatic monitoring.
Using Definiens Developer, segmentation and classification were undertaken at 2 levels to create a hierarchical image object. The first level classified an image into 5 major land cover classes. The second level brings the final classes by separating major land covers into more detailed classes. The validation was assessed by comparing result of the computer-based segmentation with a reference segmentation generated by visual interpretation. The classification accuracy was assessed by comparing the automated classification with a reference taken from aerial photos using error matrix.

Results showed that the developed rule set provided good classification result with overall classification accuracy of 83.6 % when classifying the major land cover.  Comparison of computer-based segments with a reference generated manually yielded less than 50% correspondence. Most of the differences were found as over-segmented, which means that computer-based segments provided more detailed results.
Even though the rule set gave a good classification result, it was not convincingly being robust and operational. Time inefficiency and great human interaction on modifying the rule set are still the main issues that need to be overcome in further study.

Keywords: high resolution aerial photographs, image segmentation, classification, object-oriented analysis, nature areas.

Print this activity