Unsupervised learning (clustering) of odontocete echolocation clicks
Abstract
We propose to create a method to cluster marine mammal echolocation clicks for species assemblages where little or no prior knowledge exists about these species’ signal repertoire. This will provide a tool for passive acoustic monitoring that can be used to understand how animals in a region use their habitat or how naval operations may affect their behavior. We present this proposal with a demonstration phase on data from an area where we are able to identify many species acoustically, permitting verification of the validity of the technique, and an option for applying the algorithm in a different region where the acoustics of resident species is less well understood. This method can discover echolocation click types through the application of unsupervised learning. Without any categorical knowledge of the data (unsupervised), unsupervised learning enables the partitioning of the data into similar clusters, and is one of the most effective tools for exploratory data analysis. The proposal demonstrates the potential for this the effectiveness of this technique through the application of a small subset of the proposed methods to two echolocation clicks from two odontocete species from the Southern California Bight.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 12, 2016
- Source ID
- N000141512299
Entities
People
- Marie A. Roch
Organizations
- Office of Naval Research
- Salk Institute for Biological Studies
- United States Navy