Automatic classification of humpback whale social calls

Abstract

Acoustic methods are becoming increasingly common in the study of marine mammal populations and behavior. Automating the detection and classification of whale vocalizations has been a central aim of these methods. The focus has primarily been on intra-species detection and classification, however, humpback whale (Megaptera novaeangliae) social call detection and classification has largely remained a manual task in the bioacoustics community. To automate this process, we processed spectrograms of calls using PCA-based and connected-component-based methods, and derived features from relative power in the frequency bins of these spectrograms. We then used these features to train and test a supervised Hidden Markov Model (HMM) algorithm to investigate classification feasibility.

Large-Scale Whale-Call Classification by Transfer Learning on Multi-Scale Waveforms and Time-Frequency Features

“…A large amount of the acoustic dataset titled Directional Autonomous Seafloor Acoustic Recorders (DASARs) is used to test the performance of contour tracing methods and image segmentation techniques in detecting and classifying bowhead whale calls [6]. Recently in the classification of humpback whale social calls, some researchers apply PCA-based and connected-component-based methods to derive features from relative power in the frequency bins of spectrograms and a supervised Hidden Markov Model (HMM) algorithm is then used as a classifier to investigate the classification feasibility [7]. A generalized automated detection and classification system (DCS) was developed to efficiently and accurately identify low-frequency baleen whale calls in order to tackle the large volume of acoustic data and reduce the laborious task [8].…”

Relevance-Based Feature Masking: Improving Neural Network Based Whale Classification Through Explainable Artificial Intelligence

“…In order to facilitate this process, the application of automatic classification systems, which are able to detect specific relevant patterns in the data, has been growing in popularity [3]. Examples for this include the calculation of spectrogram correlation values [9,10], extraction of frequency contours using edge detection algorithms and computation of * These authors contributed equally to this work pixel-based features [11], as well as the application of a principal component analysis to derive features from the relative power of frequency bins in spectrograms [12]. Due to the recent success of Convolutional Neural Networks (CNNs) in the fields of computer vision and natural language processing, current approaches also explore the feasibility of CNN-based models for bio-acoustic classification tasks such as whale call recognition.…”