Deep Machine Learning Techniques for the Detection and Classification of Sperm Whale Bioacoustics

Peter C Bermant,Robert J Wood,David F Gruber,Michael M Bronstein,Shane Gero

doi:10.1038/s41598-019-48909-4

Abstract

We implemented Machine Learning (ML) techniques to advance the study of sperm whale (Physeter macrocephalus) bioacoustics. This entailed employing Convolutional Neural Networks (CNNs) to construct an echolocation click detector designed to classify spectrograms generated from sperm whale acoustic data according to the presence or absence of a click. The click detector achieved 99.5% accuracy in classifying 650 spectrograms. The successful application of CNNs to clicks reveals the potential of future studies to train CNN-based architectures to extract finer-scale details from cetacean spectrograms. Long short-term memory and gated recurrent unit recurrent neural networks were trained to perform classification tasks, including (1) “coda type classification” where we obtained 97.5% accuracy in categorizing 23 coda types from a Dominica dataset containing 8,719 codas and 93.6% accuracy in categorizing 43 coda types from an Eastern Tropical Pacific (ETP) dataset with 16,995 codas; (2) “vocal clan classification” where we obtained 95.3% accuracy for two clan classes from Dominica and 93.1% for four ETP clan types; and (3) “individual whale identification” where we obtained 99.4% accuracy using two Dominica sperm whales. These results demonstrate the feasibility of applying ML to sperm whale bioacoustics and establish the validity of constructing neural networks to learn meaningful representations of whale vocalizations.

Highlights

We implemented Machine Learning (ML) techniques to advance the study of sperm whale (Physeter macrocephalus) bioacoustics
Our results show that the sperm whales’ click-based sonar and communication system is well-suited for ML-based techniques
We establish that Convolutional Neural Networks (CNNs), neural network architectures used for computer vision tasks, can successfully be used to detect sperm whale echolocation clicks from spectrograms

Summary

Introduction

We implemented Machine Learning (ML) techniques to advance the study of sperm whale (Physeter macrocephalus) bioacoustics This entailed employing Convolutional Neural Networks (CNNs) to construct an echolocation click detector designed to classify spectrograms generated from sperm whale acoustic data according to the presence or absence of a click. In contrast to the relative ease of investigating the communicative capabilities of humans and other terrestrial organisms, working with non-captive animals in the ocean environment presents numerous technical and logistical difficulties[6] Given these challenges, the development of improved computational techniques for analyzing cetacean sounds plays an important role in enabling researchers to address questions pertaining to cetacean vocal behavior. Popular among a plethora of different ANN architectures are Multi-Layer Perceptrons (MLPs) typically used for general classification and regression problems, Convolutional Neural Networks (CNNs) for image classification tasks, and Recurrent Neural Networks (RNNs) for time series sequence prediction problems

Methods

Results

Conclusion