Beluga whale acoustic signal classification using deep learning neural network models.

Area of Science:

• Marine Mammal Research
• Bioacoustics
• Artificial Intelligence in Ecology

Background:

• The Cook Inlet beluga whale (Delphinapterus leucas) population, listed as endangered in 2008, shows no signs of recovery.
• Limited ecological knowledge hinders understanding and management of threats to this declining population.

Purpose of the Study:

• To investigate beluga whale seasonal occurrence using passive acoustics.
• To develop a more efficient and accurate method for classifying beluga acoustic detections.

Main Methods:

• Deployment of passive acoustic moorings in Cook Inlet.
• Processing acoustic data with semi-automated tonal detectors and manual validation.
• Construction and application of an ensembled deep learning convolutional neural network (CNN) model for signal classification.

Main Results:

• The deep learning model achieved 96.57% precision and 92.26% recall on a testing dataset.
• The developed methodology successfully classifies beluga whale acoustic signals.
• The framework demonstrates potential for generalization to other acoustic classification tasks.

Conclusions:

• The deep learning approach significantly improves the efficiency and accuracy of beluga acoustic detection classification.
• This technology can aid in understanding beluga whale ecology and inform conservation strategies.
• The generalized framework offers a valuable tool for bioacoustic research across species.