Broadband Sonar Classification Cues: An Investigation
Abstract
The human auditory system has excellent pattern recognition capabilities and can be used to identify acoustic cues in broadband sonar classification tasks. Sonar echo-discrimination experiments were conducted with human subjects to (1) measure human auditory discrimination performance using broadband sonar echoes from geometric targets, (2) determine the acoustic cues used by subjects in discriminating between targets, (3) develop software algorithms to extract echo features similar to those used by humans, and (4) determine whether the features can be used for automatic target classification. Human subjects discriminated between a variety of target echoes, with scores exceeding 90-percent correct on most tasks, using mostly time-domain cues including time-separation pitch, rise time, and duration differences. Time- domain feature extraction software was developed, with echo features statistically representing the time-separation and amplitude of echo highlights. An automatic pattern recognition algorithm using these features obtained 90- percent correct performance on a seven-alternative material composition discrimination task with chance performance being 14-percent correct.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 01, 1986
- Accession Number
- ADA176499
Entities
People
- D. W. Martin
- W. W. Au