| Autor: |
Mahdi H. Al-Badrawi, Nicholas J. Kirsch, Christopher M. Foster, Kerri D. Seger, Yue Liang |
| Rok vydání: |
2021 |
| Předmět: |
|
| DOI: |
10.21203/rs.3.rs-831254/v1 |
| Popis: |
Climate change affects the distributions of marine mammals1, and some temperate water species are spreading northward into the Arctic Ocean2, 3. Tracking expanding species is crucial to conservation efforts and using automatic detectors and classifiers to track the locations of their vocalizations could help. Risso’s (Gg) and Pacific white-sided (Lo) dolphins were documented spreading poleward2 and make very similar sounds, making it difficult for both human analysts and classification algorithms to tell them apart. Variational Mode Decomposition (VMD) has provided both an easier visualization tool4 for human analysts and offers promising capabilities in separating call types of similar spectral and temporal properties. Here we show a new visualization tool and feature extraction technique using VMD that achieves 81.3% accuracy, even when using audio files with faint signals and high background noise levels and without context clues. Because not all dolphins whistle5–7, being able to distinguish between just their pulsed signals is important for tracking them using as many files as possible from under-sampled areas of the ocean. Automating the VMD method and expanding it to other dolphin species that have very similar pulsive signals will lead to a faster understanding of ecosystem dynamics under a changing climate than can currently be achieved. |
| Databáze: |
OpenAIRE |
| Externí odkaz: |
|
