Large-scale evaluation of automated detection and localization of bowhead whale calls in the Beaufort Sea.

Citation

Thode, A. M., D. Mathias, C. S. Nations, T. L. McDonald, and M. A. Macrander (2009). “Large-scale evaluation of automated detection and localization of bowhead whale calls in the Beaufort Sea.” In: The Journal of the Acoustical Society of America 126.4, p. 2230. DOI: 10.1121/1.3248995.

Number of Citations: 0

Abstract

An automated procedure has been developed for automated detection and localization of bowhead whale sounds with arbitrary frequency-modulated tones and applied to 2008 data collected from 41 directional autonomous seafloor acoustic recorders deployed over a 280-km swath in the Beaufort Sea. The procedure has seven sequential stages: an incoherent spectral band detector, an interval estimator, a feature extractor, a feed-forward neural network classifier, a stage to flag common calls among recorders, bearing extraction, and final localization. The three-layer neural network has ten hidden-layer units and uses 20 features extracted from the spectrogram classifying signals as either whale calls or other sounds. Manual analysis was conducted on recordings from six non-consecutive days chosen to span a variety of seismic activity levels. The first 12 h of each day yielded 141 796 true calls and 1.15$$106 other signals that were used to train the network, with 60% of that data used to adjust the weights and 40% used to evaluate convergence. The performance of both the neural network stage and the complete tracking algorithm on the remaining 12 h of manually-analyzed data from each day will be presented. [Work supported by Shell Exploration and Production Company.]