Illumination and Temperature Invariant Recognition in Multispectral Infrared Imagery

Abstract

Significant progress has been made towards achieving the research objectives in the areas of physical modeling, algorithm development, and experimental evaluation. We have carefully analyzed the physics underlying the information of airborne hyperspectral imagery over the 0.4 microns-2.5 microns spectral range which corresponds to the HYDICE and AVIRIS sensors. The new spectral radiance model includes reflected solar and scattered radiation as well as the effects of atmospheric gases and aerosols. We have shown using a statistical analysis of the radiance model that the variation in an object's spectral signature lies in a low-dimensional space. This result is the basis of a new maximum likelihood ATR algorithm that is invariant to illumination and atmospheric conditions. We have demonstrated that the new recognition algorithm significantly outperforms existing algorithms over a range of HYDICE and AVIRIS imagery over a range of conditions. We have also evaluated the use of linear models for representing mid-wave and long-wave infrared spectral reflectance functions.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Sep 01, 2000
Accession Number
ADA386225

Entities

People

  • Glenn Healey

Organizations

  • University of California, Irvine

Tags

Communities of Interest

  • Sensors

DTIC Thesaurus Topics

  • Airborne
  • Algorithms
  • Classification
  • Computer Vision
  • Earth Sciences
  • Hyperspectral Imagery
  • Identification
  • Illumination
  • Long-Wavelength Infrared Radiation
  • Materials
  • Multispectral
  • Pattern Recognition
  • Radiance
  • Radiation
  • Recognition
  • Reflectance
  • Statistical Analysis

Readers

  • Image Processing and Computer Vision.
  • Neural Network Machine Learning.
  • Spectroscopy.

Technology Areas

  • Space