Receiver-Operating-Characteristic (ROC) analysis applied to listening-test data – Measures of performance in aural classification of sonar echoes

PDF

Authors
  1. Allen, N.
Corporate Authors
Defence R&D Canada - Atlantic, Dartmouth NS (CAN)
Abstract
A Receiver-Operating-Characteristic analysis was conducted on data from two listening tests that were carried out as part of a DRDC Technology Investment Fund project on aural classification of sonar echoes. All the human listeners were DND personnel with significant experience in sonar. An automatic classifier was also tested. The results of the analysis strongly support the idea of using aural cues to discriminate between target echoes and clutter. The automatic classifier outperformed some of the human listeners and was on par with the others. Performances of the human listeners and automatic classifier in the second listening test, where a high-pass filter was applied to the echoes to remove all frequency content below 500 Hz, were definitely poorer than for the first test, where the full available bandwidth of 0 - 2 kHz was exploited, but still substantially better than chance performance.

Il y a un résumé en français ici.

Keywords
measures of performance;human-performance testing;ROC analysis;aural classification;sonar echo classification;Human performance
Report Number
DRDC-ATLANTIC-TM-2007-353 — Technical Memorandum
Date of publication
01 Aug 2008
Number of Pages
78
DSTKIM No
CA031918
CANDIS No
530957
Format(s):
CD ROM

Permanent link

Document 1 of 1

Date modified: