Skip to content

lifewatch/pyporcc

Repository files navigation

PyPorCC

PyPorCC is a package that allows the detection and classification of Harbor Porpoises' clicks. The detection of clicks in continuous files is a python adaptation of the PAMGuard click detector algorithm.

Gillespie D, Gordon J, McHugh R, McLaren D, Mellinger DK, Redmond P, Thode A, Trinder P, Deng XY (2008) PAMGUARD: Semiautomated, open source software for real-time acoustic detection and localisation of cetaceans. Proceedings of the Institute of Acoustics 30:54–62.

The classification is done using the PorCC algorithm, adapted to python from the paper:

Cosentino, M., Guarato, F., Tougaard, J., Nairn, D., Jackson, J. C., & Windmill, J. F. C. (2019). Porpoise click classifier (PorCC): A high-accuracy classifier to study harbour porpoises ( Phocoena phocoena ) in the wild. The Journal of the Acoustical Society of America, 145(6), 3427–3434. https://doi.org/10.1121/1.5110908

Also other models can be trained. The implemented ones so far are:

  • svc: Support Vector Machines
  • lsvc: Linear Support Vector Machines
  • RandomForest: Random Forest
  • knn: K-Nearest Neighbor

Usage

Click detector

The Click detector can be used in continuous wav files (with higher than 300 kHz sampling rate) or in the SoundTrap HF output files (*.bcl + *.dwv). For SoundTrapHF files, you can create a ClickDetectorSoundTrapHF object with the necessary parameters and run it as:

import pathlib
import pyhydrophone as pyhy

from pyporcc import ClickDetectorSoundTrapHF, ClickDetector, PorCC, Filter


sound_folder = pathlib.Path("./../tests/test_data/soundtrap")
save_folder = pathlib.Path('./../tests/test_data/output')

# Hydrophone
model = 'ST300HF'
name = 'SoundTrap'
serial_number = 67416073
soundtrap = pyhy.soundtrap.SoundTrapHF(name=name, model=model, serial_number=serial_number)

# Filters parameters
lowcutfreq = 100e3              # Lowcut frequency
highcutfreq = 160e3             # Highcut frequency

# Define the filters
pfilter = Filter(filter_name='butter', filter_type='bandpass', order=4,
                                frequencies=[lowcutfreq, highcutfreq])
dfilter = Filter(filter_name='butter', filter_type='high', order=4, frequencies=20000)
classifier = PorCC(load_type='manual', config_file='default')

cd = ClickDetectorSoundTrapHF(hydrophone=soundtrap, save_folder=save_folder, convert=True,
                              classifier=classifier, prefilter=pfilter, save_noise=False)
cd.detect_click_clips_folder(sound_folder, blocksize=60 * 576000)

For continuous data, just make sure you use the class ClickDetector object instead of a ClickDetectorSoundTrapHF! The rest of the code should be the same (except the hydrophone definition, which will depend on the instrument you use)

Note

Please note, the clicks PAMGuard's Click Classifier classified as porpoise clicks appear as 0 in both ClassifiedAs and ManualAssign fields.

Citation

DOI