delphinID North East Atlantic

dolphin
delphinID classifiers (i.e. classification models) for the Northeast Atlantic - process clicks and whistle contours detected using PAMGuard’s Whistle & Moan Detector and Click Detector to classify seven delphinid species.

Introduction

delphinID models are deep convolutional neural networks (CNNs) trained in Python Tensorflow 2.18.0 to automatically classify detections of delphinid whistles and clicks to species based on the distributions acoustic frequencies in groups of whistles or clicks detected using PAMGuard’s Whistle & Moan Detector (WMD) and Click Detector, respectively. delphinID click and whistle models for the Northeast Atlantic were trained using groups of detections within 4-second time windows, while the output of these models is used as a feature vector for additional Random Forest layer which provides a recording-level species classification based on acoustic information from both whistles and clicks. As the detection-based input for classification is computationally inexpensive to compute, delphinID can easily run in PAMGuard at up speeds up to 64x real-time. Both click and whistle classifiers were trained and evaluated exclusively on visually confirmed single species recordings of these species made in the Northeast Atlantic and North Sea and only evaluated on recordings spatiotemporally independent from those used to train the models.  

Cross-validated testing estimates delphinID to classify events with average accuracy ranging from 80% for Delphinus delphis to 92% for Lagenorhynchus albirostris.

delphinID running in PAMGuard in real time using both a click and whistle model

Cross-validated testing estimates delphinID to classify recordings with an average accuracy of 86%, and ranges from 80% for Delphinus delphis to 92% for Lagenorhynchus albirostris. Species that can be classified by the Northeast Atlantic delphinID models are:

  1. Atlantic white-sided dolphin (Lagenorhynchus acutus)

  2. Common bottlenose dolphin (Tursiops truncatus)

  3. Killer whale (Orcinus orca)

  4. Long-finned pilot whale (Globicephala melas)

  5. Risso’s dolphin (Grampus griseus)

  6. Short-beaked common dolphin (Delphinus delphis)

  7. White-beaked dolphin (Lagenorhynchus albirostris)

Reference

Paper in prep.

Model

Northeast Atlantic delphinID click and whistle classifiers can be found on Zenodo.DOI

delphinID does not use raw audio data as its input for classification; instead, it processes detected PAMGuard whistle contours and click detections. The first step is therefore to run the WMD and the Click Detector on wav files in PAMGuard. The output is then run through the delphinID modelsloaded into the Deep Learning module in PAMGuard. The resulting click and whistle predictions can then be combined for final species classification via an R application (see README file) also accessible through the Zenodo download. Also included in the download are a PAMGuard settings file (.psfx) for delphinID and a README file containing documentation, tutorials, and extra information. 

To run delphinID, users must set the location of the database, binary store and sound files, and open both of PAMGuard’s deep learning modules. To do this go to Settings -> Raw Deep Learning Classifier_Clicks and select the deep learning models on their local machine. The configuration will then detect whistle contours and clicks from raw sound data, pass the detections to both deep learning modules and save the results in the PAMGuard database. Also note that delphinID can run in viewer mode if whistles and/or click detections are part of the data model.