From massive detections and localisations of orca at orcalab over three years to real-time survey joint to environmental conditions
Résumé
Orcas (Orcinus orca) can produce 3 types of signals:clicks, whistles and vocalizations. This study focuses onorca vocalizations from northern Vancouver Island (Han-son Island) where the NGO Orcalab developed a multihy-drophone recording station to study orcas. The acousticstation is composed of 5 hydrophones and extends over 50km2of ocean. Since 2015 our team is continuously stream-ing the hydrophone signals, yielding nearly 50 TB of syn-chronous multichannel recordings. We trained a Convolu-tional Neural Network (CNN) to detect orca vocalizations,by transfer learning from a bird activity dataset. We re-late the detected presence and absence of the orca vocal-izations to the time of the day, the moon phases and thetides. This real-time detector opens new insight on thesestreamed submarine recording, allowing online automaticsurvey of the orca in the whole area. It thus assess in whicharea the orcas are present. It then could be used to monitorthe effect of traffic pressure and anthropophonic pollutionon this species. We finally discuss towards the installationof small quadraphonic system acoustic device to completeand detail these observations from Orcalab, in order to getmore precise relative positions of the orcas.
Origine : Accord explicite pour ce dépôt