Deteclic: A multi-method detector of sperm whale's click

Abstract

In the French EEZ of Crozet and Kerguelen Islands, longline fishing is heavily impacted by the depredation of sperm whale, with significant socio-economic implication and ecological impact. Until now, only visual observations of the surface have been used to monitor this phenomenon. Their limitations have raised the need to extend the methods of investigation. Thus, an autonomous acoustic recorder was attached directly to the longline and several hours of recording were collected, disqualifying de facto manual annotation of the data. To overcome this difficulty, we developed an automatic detector, based on five simple methods from the literature: intercorrelation with a reference signal, a Teager-Kaiser energy operator convolved with a Gabor function, spectrogram analysis, kurtosis-based statistical detection and analysis of the Daubechies 15 wavelet-transform. The different methods runs independently with a click-length analysis window allowing a click-by-click detection, and then performs a vote to limit the false alarm rate. Our tool performance is assessed on a dataset where 2450 clicks have been identified by an expert. The clicks detected give valuable indication of the presence/absence of cetacean around the longline, the level of click detection allows us to dissociate the different acoustic behaviours leading to the detection of the depredation event.