Automatic Classification of Biological Sounds in the Arctic.

Ambient underwater recordings in the Arctic are generated by a complex mixture of physical processes and biological events. Even for experts, it is difficult and time-consuming to detect and identify biological transients. During this project, improved methods for reviewing multichannel acoustic dat...

Full description

Bibliographic Details
Main Author: Fristrup, Kurt M.
Other Authors: WOODS HOLE OCEANOGRAPHIC INSTITUTION MA
Format: Text
Language:English
Published: 1996
Subjects:
Biological Oceanography
Acoustic Detection and Detectors
*ACOUSTIC DATA
*AQUATIC ANIMALS
TRANSIENTS
TWO DIMENSIONAL
CLASSIFICATION
RECORDING SYSTEMS
COLLECTION
SONAR SIGNALS
UNDERWATER
ARCTIC REGIONS
MARINE BIOLOGICAL NOISE
MULTICHANNEL
Arctic
Online Access:http://www.dtic.mil/docs/citations/ADA329473
http://oai.dtic.mil/oai/oai?&verb=getRecord&metadataPrefix=html&identifier=ADA329473
Description
Summary:Ambient underwater recordings in the Arctic are generated by a complex mixture of physical processes and biological events. Even for experts, it is difficult and time-consuming to detect and identify biological transients. During this project, improved methods for reviewing multichannel acoustic data and promising techniques for automatic classification of biological sounds were developed. Two analytical methods demonstrated the promise of automatic recognition for these sounds. The first technique was a Classification Tree. This method produced a classifier consisting of a sequence of simple rules based on individual features. A classification tree was computed that divided the collection of sounds into 23 categories; these 22 rules were sufficient to correctly identify 591 of 699 sounds to species, or about 85% correct classification. In addition to the classification tree, a principal component analysis was also conducted on these data. Principal component scores were extracted from the rescaled data, to obtain new features that were mutually orthogonal, and identify which axes expressed the preponderance of the overall variation. The dominant principal component scores were then subjected to a discriminant function analysis, to obtain a set of two-dimensional projections that provide a useful perspective on the distinctiveness of the species' sounds.

Similar Items