The wide-band detection and classification system
Abstract
Detection and classification of marine mammals using passive acoustic monitoring is traditionally conducted on a species-by-species basis separate algorithms for each species characteristic calls. When an application requires the detection of many species simultaneously, this approach becomes inefficient at best, and impractical at worst. Some detection and classification systems seek to detect a class of calls, such as tonal or pulse calls. The low-frequency detection and classification system (LFDCS) is one such system that uses (1) pitch tracking to detect and characterize all tonal sounds in a recording or real-time audio stream, and (2) a separate discriminant function analysis to classify sounds by call type and species. This allows the detection and classification of many species simultaneously in a single processing run. The LFDCS, like all tonal marine mammal detectors, relies on spectrograms produced with the short-time Fourier transform (STFT). These spectrograms characterize spectral content on a linear frequency scale. This limits the band of frequencies that can be effectively monitored to roughly 4 octaves, yet marine mammals make tonal and pulse sounds over a range of 12+ octaves. I propose to implement an efficient algorithm to create spectrograms based on the constant-Q transform (CQT), a technique to estimate spectral content on a logarithmic frequency scale, and to incorporate CQT-based spectrograms in a new wide-band detection and classification system (WBDCS). The WBDCS will also incorporate a detection and classification system for pulse sounds (e.g., echolocation clicks, minke whale pulse trains, walrus knocks) that will take advantage of the wide-band processing required for the creation of CQT-based spectrograms. This new system will be conceptually and practically similar to the LFDCS, except the monitored frequencies will extend from 14 Hz to ~50 kHz, the frequency resolution will be more appropriate for pitch tracking tonal calls throughout this frequency range, and time-domain front-end processing will be used to detect and classify pulse calls. This single system will be capable of simultaneously detecting tonal and pulse calls ranging from the low-frequency moans of blue whales to the high-frequency clicks of beaked whales. The WBDCS will be implemented on both a desktop computer for processing archived recordings and on the DMON/DMON2 instrument for in-situ real-time detection and classification from autonomous platforms.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 04, 2018
- Source ID
- N000141812810
Entities
People
- Mark F. Baumgartner
Organizations
- Office of Naval Research
- United States Navy
- Woods Hole Oceanographic Institution