Zaugg, S., van der Schaar, M., Houégnigan, L., André, M.
Real time classification system for acoustic events in the marine environmen
4th International Workshop on Detection, Classification and Localization of Marine Mammals Using Passive Acoustics, Sep 2009

Resum:
The automated, real time classification of acoustic events in the marine environment is an important tool to study how anthropogenic sound pollution affects marine mammals and for mitigating human activities that are potentially harmful to them. However the classification in a fully automated way is challenging due to the diversity of acoustic events and background noises typically found. We present a real time classification system targeted at many important groups of acoustic events (clicks, buzzes, calls, whistles from several cetacean species, tonal and impulsive shipping noise and explosions). The system is composed of two stages: The first stage, made of several detection algorithms, detects segments that contain acoustic events and tags them according to broad classes (e.g. low frequency impulses, ultrasonic impulses, short tonals). The second stage, made of several classification algorithms, classifies events that have been detected in the first stage into more specific classes, which have practical relevance (e.g. impulsive ship noise, ultrasonic cetacean clicks, cetacean buzzes, whistles). The system’s reliability was tested on data from several sites, including the datasets made available for the workshop. The achieved classification performance indicates that the system is reliable to automatically classify acoustic events in real time. This system will be useful to pre-process the very large data volume that can be gathered during long term PAM campaigns or to detect the presence of cetaceans in real time for mitigation.

Projecte: LIDO, Listening to the Deep-Ocean Environment

Projecte: ESONET, Network of Excellence