Classification of Terrestrial Materials and Vegetation Using Remotely Sensed Multi-Spectral Data at the Atlantic Undersea Test and Evaluation Center (AUTEC) Main Base on Andros Island, Bahamas

Abstract

The classification of vegetation and materials--both natural and man-made--in the terrestrial environment was conducted using high-spatial-resolution, multi-spectral satellite imagery obtained from the IKONOS-2 sensor. The results of three supervised classification techniques, the Maximum Likelihood Classifier (MLC), the Spectral Angle Mapper (SAM\) classifier, and Mahalanobis Distance classifier, are presented. Ground truth data were used to compare the statistical accuracy of the different classifier techniques to determine which classifier provided the best overall results. Based on the results of the statistical comparison, producer accuracy, omission error, and commission error, it was determined that the MLC provided the best overall classification. This technique was then optimized using the training data sets, and the process was implemented over the entire area of the satellite image. One site from within the image was selected for a final ground truth comparison to determine the overall improvement of the optimized classifier.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Jan 25, 2005
Accession Number
ADA636487

Entities

People

  • Marc D. Ciminello
  • Thomas K. Szlyk

Organizations

  • Naval Undersea Warfare Center

Tags

Communities of Interest

  • Space

DTIC Thesaurus Topics

  • Accuracy
  • Department Of Defense
  • Detectors
  • Environment
  • Environmental Protection
  • Errors
  • Images
  • Machine Learning
  • Materials
  • Natural Resources
  • Photographs
  • Satellite Imaging
  • Scattering
  • Spectra
  • Supervised Machine Learning
  • Test And Evaluation
  • Undersea Warfare

Readers

  • Atmospheric Remote Sensing.
  • Neural Network Machine Learning.
  • Regression Analysis.

Technology Areas

  • Space