Comparative evaluation of the Mercury classification algorithm – On the influence of the number of bands on classification accuracy using hyperspectral data

PDF

Authors
  1. van Chestein, Y.
Corporate Authors
Defence R&D Canada - Valcartier, Valcartier QUE (CAN)
Abstract
As part of the HYMEX technology demonstration project, DRDC Valcartier asked Dr. Derek Peddle, of the University of Lethbridge (U of L), to analyse a 15 m spatial resolution Probe-1 airborne hyperspectral image of CFB-Gagetown, New-Brunswick and 4 m resolution AISA hyperspectral imagery of CFB-Wainwright, Alberta. Among the work performed was the classification of the data in seven classes, conifer, deciduous, mixed, grass, clearcut areas, dirt road and water. U of L used its own Mercury algorithm to perform the classification, along with the Maximum Likelihood algorithm of ITT’s ENVI software package. The university was then able to compare both algorithms as to classification accuracy. We endeavoured to validate the results obtained by U of L and expand the comparison to all of ENVI’s tools. We found that Mercury consistently gives results at least as good as some of ENVI’s tools and gives much better results than the other tools, and that with fewer limitations. This expanded comparison serves to validate the decision to fund the work performed by U of L and the acquisition of the Mercury algorithm for its integration into DRDC’s Hyperspectral Operational Support Tools (HOST), developed under HYMEX.

Il y a un résumé en français ici.

Report Number
DRDC-VALCARTIER-TM-2010-385 — Technical Memorandum
Date of publication
01 Mar 2011
Number of Pages
66
DSTKIM No
CA035223
CANDIS No
534829
Format(s):
Electronic Document(PDF)

Permanent link

Document 1 of 1

Date modified: