Distinguishing FM‐modulated bowhead whale calls from airgun and other biologic signals using image processing, feature extraction, and neural networks.

Abstract

The detection and identification of repetitive or stereotyped bioacoustic signals in raw acoustic data are typically accomplished via matched filtering or spectrogram correlation techniques. However, the detection of FM‐modulated sweeps that have variable frequency‐modulated contours or bandwidths is still an active challenge. The problem is made more complex by the possible presence of other frequency‐modulated signals such as airguns or other biologics. To provide a specific example, in 2007 and 2008 directional autonomous recording packages (DASARs) were deployed in the Beaufort Sea to monitor the annual migration of the bowhead whale (Balaena mysticetus) population during seismic exploration activities. A portion of these data sets has been manually analyzed, providing an opportunity to characterize the miss and false alarm rates of various FM‐modulated detection and classification methods. Here literature and software on bioacoustic FM‐contour tracing and classification are reviewed, with a particula...