Transient Classifier Systems and Man-Machine Interface Research.
Abstract
The results of the experiment showed that transient detection and classification performance are highly independent, and both are very sensitive to signal-to-noise ratio (SNR). Unknown transients were recognized rapidly; however, performance at low SNR was not comparable to that against known transients. Transient specific syntax proved to be an even stronger determinant of performance than the known vs. unknown condition. Novice performance in detecting a target by its transient emissions was comparable to theoretical best current broadband techniques. Experienced sonar operators outperformed the novices by 12 dB. The automatic classification algorithm research demonstrated use of syntactic and semantic state variable feature-space representations to perform computationally efficient classification of transient patterns (50 times real-time in FORTRAN) and large-scale reduction of data (500:1). The algorithm recognized many singular and correlated transient events. An unexpected and exciting result was recognition and modal separation of mixed mode tonal signals as correlated transients in the time domain.
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 31, 1987
- Accession Number
- ADA186213
Entities
People
- Richard Kram
- Robert Sax